HbA1c-based personalization of glucose management in critical illness: a target trial emulationABSTRACT
Objective: To evaluate the effect of HbA1c-guided glycaemic control on 30-day intensive care unit mortality in intensive care unit patients.
Design: Target trial emulation using the parametric g-formula. Setting: A single tertiary academic hospital intensive care unit in Melbourne, Australia. Patients: All adults admitted between January 2019 and December 2023 with an admission HbA1c measurement and complete intensive care unit admission/discharge data. Interventions: Three glucose control strategies were compared: natural course (usual care), NICE-SUGAR target (8 - 10mmol/L), and personalised HbA1c-guided target (usual glycaemia ± 0.8mmol/L).
Results: Among 3,541 patients (64% male, median age 68 years), mortality was 6.46% (95%CI 4.16% - 8.83%) under the natural course, 6.52% (95%CI 4.40% - 8.82%) with NICE-SUGAR, and 6.56% (95%CI 4.40% - 8.85%) with HbA1c-guided targets. Absolute risk differences versus usual care was 0.10% (95%CI -0.26% - 0.93) for HbA1c-guided strategy. Subgroup analyses by diabetes status, admission type, illness severity, and septic shock showed no survival benefit.
Conclusion: In this large single-centre target trial emulation, personalising glycaemic targets based on admission HbA1c did not improve 30-day intensive care unit mortality compared with either usual care or NICE-SUGAR targets. These findings suggest that HbA1c-guided glucose control is unlikely to confer meaningful survival benefit under current implementation approaches.
Keywords:
Intensive care unit; Glucose management; Target trial emulation, Critical care
INTRODUCTION
Hyperglycaemia is common in critically ill patients and is associated with worse outcomes, including infection, impaired wound healing, and mortality.(1) While early trials suggested benefit from intensive glycaemic control, subsequent large multicentre trials, including Normoglycemia in Intensive Care Evaluation-Survival Using Glucose Algorithm Regulation (NICE-SUGAR), showed harm or no benefit, particularly due to increased hypoglycaemia.(2) As a result, current guidelines recommend a conventional target of < 10mmol/L (180mg/dL),(3) but despite its well-established pathophysiology, the optimal approach to glycaemic control in the intensive care unit (ICU) remains controversial, particularly regarding whether treatment targets should be universal or personalized.(4)
A universal glycaemic target may not suit all patients. Observational data suggest that the optimal glucose level may differ based on chronic glycaemic status, with patients with diabetes tolerating higher glucose levels.(5) Hemoglobin A1c (HbA1c), a measure of long-term glycaemic exposure, reflects mean glucose levels over approximately 3 months and is a key biomarker in diabetes diagnosis and prognosis.(6) Its availability at hospital or ICU admission offers a potential opportunity to personalise glycaemic targets according to a patient's usual glycaemia, rather than applying a fixed threshold to all.
The CONTROLING trial tested this hypothesis by randomizing ICU patients to either conventional control (< 10mmol/L [180mg/dL]) or personalized control based on HbA1c.(7) Although the trial found no overall mortality benefit, evidence from selected ICU populations suggests potential benefit of personalizing glycaemic targets.(8-10) In cardiac surgical and neurosurgical cohorts, tighter or individualized control has been associated with improved outcomes.(8-10) Moreover, observational data demonstrate complex interactions between chronic and acute glycaemia.(11) For instance, the glycaemic level associated with lowest mortality appears to be higher in patients with diabetes than in those without, implying that a universal glucose target might inadvertently overtreat some patients while undertreating others.(12) These findings raise the question of whether specific ICU sub-populations might still benefit from a personalized approach, a hypothesis that can now be evaluated through robust target trial emulations using real-world data.
The causal research question addressed by this target trial emulation is: among critically ill adult patients admitted to the ICU, what is the effect of maintaining glucose trajectories within personalized HbA1c-based targets compared with conventional glucose control strategies, including NICE-SUGAR targets and observed clinical practice, on ICU mortality within 30 days of ICU admission? Accordingly, the primary objective of this study was to evaluate the effect of HbA1c-guided glycaemic control on 30-day intensive care unit mortality in intensive care unit patients.
METHODS
Study design
We conducted a target trial emulation assessing the effect of personalized glycaemic control based on HbA1c, on clinical outcomes of critically ill patients. Details of the trial emulation protocol can be found in table 1. A target trial emulation was chosen in place of a traditional observational study to reduce the effect of selection/immortal time bias and allow stronger causal inferences to be made.(13) By analyzing retrospective data via a target trial emulation design we are also able to utilize existing records to provide faster insight compared to a standard clinical trial. This study was approved by the Austin Hospital Human Research Ethics Committee (VicTRI-16351-HREC/111269/Austin-2024), with a waiver of informed consent.
Summary of the protocol of the target trial assessing the impact of different glycaemic control strategies on mortality
Patients
All adult patients (≥ 18 years old) admitted to the ICU of the Austin Hospital, Melbourne, were eligible for inclusion. If a patient had multiple admissions, only the first admission was considered for inclusion. In addition, only patients with an HbA1c measure at hospital admission were included in the analysis. At our institution, HbA1c testing is supported by an institutional protocol embedded in the electronic medical record (EMR). Since approximately mid-2013, HbA1c has been routinely measured in hospitalized patients aged ≥ 54 years if no HbA1c result was available within the preceding 3 months. This automated testing protocol facilitates systematic assessment of chronic glycaemic status and reduces reliance on clinician-directed testing alone. Patients were excluded only if their data in ICU admission or discharge date and time was missing.
Data collection
All patient baseline and outcome data were drawn from the Australian and New Zealand Intensive Care Society (ANZIC) Adult ICU Patient Database (APD) and Austin EMR. Data collected included patient demographics, anthropometric measurements, admission category, comorbidities including diabetic status, organ support, and clinical outcomes. Insulin dosage and administration, and glucose measurements were drawn from the EMR. All available measurements (from pathology, blood gas and capillary) were retrieved. All data was timestamped. The HbA1c at hospital admission was used to calculate the usual glycaemia for each patient using the formula below:
Intervention
In the HbA1c strategy group, the glycaemia target was set to usual glycaemia (calculated based on the formula above) ± 0.8mmol/L.(7) In the usual care group (natural course), the actual glycaemia achieved during ICU stay was considered. Finally, in the NICE-SUGAR strategy, the glycaemia target was set to 8 to 10mmol/L.(2)
Outcomes and study endpoints
The primary outcome of interest was 30-day ICU mortality. Discharge from ICU was considered a competing event because it precluded the later occurrence of the primary outcome. Patients were followed from the ICU admission (Day Zero) until either death, ICU discharge, or Day 30 in the ICU, whichever occurred first. In all analyses, time zero was the ICU admission and the maximum follow-up time was 30 days from ICU admission. The 30-day period was considered because more than 99% of the patients had less than 30 days of length of stay. The follow-up was divided in blocks of 6 hours.
Analysis plan
Maintenance of glycaemia within proposed targets is a time-dependent exposure. In addition, there are time-dependent confounders affected by prior treatment that also predict future glucose levels and future outcome, conditional on past treatment. The use of standard regression methods in this situation results in biased estimates. Target trial emulation using parametric g-formula is an accepted statistical option to address these points when comparing dynamic treatment regimens.(14)
Target trial emulation creates a framework for causal inference from observational data. It explicitly emulates the components of a clinical trial. It includes screening patients and identifying patients eligible for treatment allocation based on their baseline characteristics and inclusion criteria for the intervention (e.g., ICU admission). It follows patients from "randomization emulation time" (time zero) to the end of planned follow-up and conducts the same analysis as would be conducted for the corresponding target trial.(13)
The interventions were defined at the level of glucose trajectories rather than specific treatment actions. At each 6-hour interval, the simulated intervention enforced glucose values within the predefined target range corresponding to each strategy. This was implemented using the parametric g-formula by replacing observed glucose values with values sampled from the conditional distribution of glucose consistent with the intervention strategy and covariate history. This approach models the effect of maintaining glucose exposure within the target range, regardless of the specific clinical actions required to achieve that target, such as insulin administration or nutritional adjustments. Accordingly, the causal estimand corresponds to the per-protocol effect of maintaining glucose trajectories within the specified target range over time under full adherence, rather than the effect of specific treatment interventions used to achieve those glucose levels. The percentage of moments when an intervention was applied reflects the proportion of timepoints at which the observed glucose value fell outside the target range and was therefore replaced by a value consistent with the simulated intervention. When multiple glucose measurements were available within a 6-hour block, the mean glucose value was used to represent glycaemic exposure during that interval. The natural course represents the observed glucose trajectories and management patterns under routine clinical practice, without modification. Within the parametric g-formula framework, the natural course reflects the empirical joint distribution of glucose levels, treatments, and patient characteristics as observed in the data.
Statistical analysis
All continuous data are reported as medians (quartile 25% - quartile 75%) and categorical data as numbers and percentages. The primary analysis was conducted using parametric g-formula. Baseline confounders included in the models were: age, sex, body mass index, admission diagnosis, presence of septic shock (according to Sepsis-3 criteria), Australian and New Zealand Risk of Death (ANZROD), usual glycaemia at hospital admission (based on HbA1c and calculated using the formula described above), total Charlson co-morbidity score, presence of diabetes, presence of chronic pulmonary disease, congestive heart failure, chronic kidney disease, highest lactate at ICU admission, use of vasopressor and/or inotropes, invasive ventilation, renal replacement therapy and Sequential Organ Failure Assessment (SOFA) score at ICU admission. Time-dependent covariates included creatinine levels, pH, mean arterial pressure, use of any oral or subcutaneous diabetic medication, use of any type of insulin and the glucose levels. All models included the hour (in blocks of 6 hours), and the time-dependent intercept was estimated by a smooth function of the day since the beginning of follow-up using natural cubic splines with five knots.
Missing baseline data from the first 24 hours after ICU admission were imputed using multiple imputation by chained equations (Table 1S - Supplementary Material). The list of predictors included the outcome variable, failure time, baseline covariates and laboratory tests. In the case of missing covariate values during the follow-up, the last observed value of a covariate was carried forward, reflecting clinical practice at the bedside where treatment decisions are based on the most recent available measurement when no new result has been obtained. In our single-centre ICU, physiological and laboratory measurements are obtained frequently, particularly during clinical deterioration, limiting prolonged reliance on carried-forward values. Subgroup analyses were performed according to diabetes (yes versus no), severity of illness (Acute Physiology and Chronic Health Evaluation [APACHE] III ≥ 62 versus < 62), type of admission (medical versus surgical), presence of septic shock (yes versus no), presence of sepsis (yes versus no), cardiac surgery (yes versus no), major surgery or trauma (yes versus no), and neurological admission (yes versus no).
As the g-formula estimates marginal risk measures, risk ratios (RR), rather than hazard ratios, are reported because the model focuses on absolute differences in risk over time rather than instantaneous rates of events. The causal contrast compares the per-protocol effect of maintaining glucose trajectories within each predefined target range with the natural course, defined as the observed glucose trajectories under routine clinical care. All analyses were performed in R Version 4.3.3 (R Foundation for Statistical Computing, Vienna, Austria).
RESULTS
Patients
From January 2019 until December 2023, 64,025 patients were screened for eligibility. After exclusions, 3,541 patients met the inclusion criteria and were included in the present study (Figure 1S - Supplementary Material). Most of the patients were male (64%) with a median age of 68 (59 - 76) years old and with a predicted risk of death of 5% (1% - 15%) (Table 2). Half of the patients were admitted due to surgical reasons (47%), with a median SOFA of 5 (3 - 7), 16% had sepsis, 62% received vasopressors, and 48% received mechanical ventilation. Diabetes was present in 39% of the patients and, at admission, the median HbA1c was 5.9% (5.4% - 6.6%) and the median glucose was 8.7 (7.2 - 10.6) mmol/L (Table 2). Additional baseline characteristics, including admission category and hospital source of admission are reported in table 2S (Supplementary Material).
Pathology test and medications
At ICU admission, median creatinine was 92.6 (70.6 - 138.5) µmol/L, pH was 7.38 (7.35 - 7.41), and median lactate was 2.2 (1.5 - 3.2) mmol/L (Table 3S - Supplementary Material). During ICU stay, mean, highest and lowest glucose levels were 8.7 (7.2 - 11.1), 9.5 (7.5 - 13.1), and 7.8 (6.4 - 9.4) mmol/L, respectively. The glucose levels over ICU stay are reported in figure 1.
At ICU admission, 0.1% of the patients were using any non-insulin glucose-lowering medications, and 2.3% were receiving insulin, mainly rapid and short acting, and with a median dose of 8 (6 - 10) units (Table 4S - Supplementary Material). During ICU stay, 4.1% of the patients received a non-insulin glucose-lowering medications with a median time until first dose of 12 (9 - 56) hours, and 22% received insulin with a median time until first dose of 24 (12 - 57) hours, and a median dose of 8 (4 - 10) units (Table 4S - Supplementary Material).
Clinical outcomes
Clinical outcomes are reported in table 2 and table 5S (Supplementary Material). The incidence of hypoglycaemia and hyperglycaemia during ICU stay was 0.3% and 44%, respectively. At Day 30, 6.1% of the patients had died (median time until mortality 4 [2 - 8] days) (Figure 2S - Supplementary Material) and 93% of the patients were discharged alive at a median time of 2 (1 - 4) days.
Intervention applied on glucose levels
The simulated trajectories are presented in figure 1 and characteristics of the intervention are reported in table 6S (Supplementary Material). Blood sugar levels remained close to the proposed range in each of the defined strategies, and a good treatment separation was achieved (Figure 1 and table 6S [Supplementary Material]). The percentage of moments when an intervention was applied to correct the glucose levels was higher in the HbA1c group than in the NICE-SUGAR group (64% versus 57%). Mean glucose levels during treatment period in the natural course, HbA1c, and NICE-SUGAR groups were 10.9 ± 4.8mmol/L, 7.7 ± 2.7mmol/L, and 9.1 ± 0.8mmol/L, respectively. As expected, there was a higher variation of glucose levels during follow-up in the natural course than in the HbA1c, and NICE-SUGAR group (44% versus 35% versus 9%, respectively), confirming the correct application of the proposed interventions, with good adherence to the treatment (Figure 1 and table 6S [Supplementary Material]).
Impact of blood glucose control strategy on outcome
In this target trial emulation assessing the impact of different glucose control strategies on 30-day ICU mortality, the estimated mortality under the natural course was 6.46% (95% confidence interval [95%CI] 4.16% - 8.83%). When applying the HbA1c-based strategy, mortality was 6.56% (95%CI 4.40% - 8.85%), corresponding to an absolute difference of 0.10% (95%CI -0.26% - 0.93%) and a RR of 1.02 (95%CI 0.96 - 1.16) compared to the natural course (Figure 2 and Table 7S [Supplementary Material]). Similarly, the NICE-SUGAR strategy yielded a 30-day ICU mortality of 6.52% (95%CI 4.40% - 8.82%), with an absolute difference of 0.07% (95%CI -0.15% - 0.83%) and a RR of 1.01 (95%CI 0.98 - 1.15) compared to the natural course (Figure 2 and Table 7S [Supplementary Material]). Comparable parametric versus nonparametric estimates of the covariate means were reported, suggesting correct model specification for the parametric g-formula (Figure 3S - Supplementary Material).
Estimated 30-day intensive care unit mortality risk under different glucose control strategies.
Subgroup analyses are reported in figure 3 and table 7S (Supplementary Material) and the findings were consistent across all subgroups, including in patients with diabetes. Because some subgroups contained relatively few mortality events, the corresponding model-based estimates show wider confidence intervals and should be interpreted cautiously. Glucose levels and simulated trajectories in patients with diabetes are presented figure 4S (Supplementary Material).
Subgroup analyses of 30-day intensive care unit mortality under different glucose control strategies.
DISCUSSION
Key findings
In this target trial emulation using granular ICU data and the parametric g-formula, we assessed the effect of a personalized glucose management strategy, based on admission HbA1c, on 30-day ICU mortality among critically ill patients. We found no significant difference in mortality when comparing the HbA1c strategy to either usual care or the NICE-SUGAR strategy. This finding was consistent across all prespecified subgroups, including patients with and without diabetes.
Relationship to previous studies
Our findings align with the results of the CONTROLING trial, the largest clinical trial to date examining individualized glycaemic targets in the ICU. That study also reported no survival benefit of personalizing blood glucose targets based on HbA1c.(7) However, CONTROLING was limited by a delay of up to 96 hours between ICU admission and randomization, during which patients in the personalized arm received conventional glucose control. The modest separation in achieved glucose levels between study arms, and only 51% time-in-target-range in the personalized group, further diluted the potential contrast between strategies. While CONTROLING found higher mortality among non-diabetic and surgical patients in the personalized arm,(7) we did not observe statistically significant subgroup differences; nonetheless, these same groups also trended toward the highest risk in our study.
Calls for high-quality research in this area are ongoing. A recent 2024 Society of Critical Care Medicine (SCCM) systematic review identified the lack of clinical trials evaluating personalized glycaemic targets as a major evidence gap.(3) Prior work has highlighted the dynamic metabolic shifts in critical illness and the theoretical benefits of matching insulin therapy to an individual's baseline glycaemic physiology.(3,15,16) However, translating these insights into clinical benefit has proven difficult.
Despite decades of research, the question of whether glucose control in the ICU should be individualized remains unresolved. While our study does not support mortality benefit with personalized targets, it also does not preclude benefits in other outcomes such as glycaemic variability, functional recovery, or long-term health status. The implementation of continuous glucose monitoring and automated insulin delivery systems may offer new opportunities to improve target adherence and safety, potentially realizing the vision of an ‘artificial pancreas’ in critical care. Future studies should also consider stratified enrolment by diabetic status and illness severity to better assess whether particular subgroups may derive benefit.
Although mortality was similar across strategies, the intervention strategies produced meaningful differences in glucose exposure and variability. These findings suggest that personalized glycaemic targets can alter glucose trajectories without necessarily translating into measurable survival benefits. From a clinical perspective, this supports the concept that moderate glycaemic control, as currently practiced, already achieves glucose levels within a physiologically acceptable range for most patients. More aggressive or personalized targeting may improve biochemical metrics such as glucose variability or time-in-range but may not substantially influence survival outcomes in the absence of other mechanistic benefits.
Implications of study findings
Our findings suggest that personalizing glycaemic control based on HbA1c does not confer a mortality benefit compared to standard approaches. Both the HbA1c and NICE-SUGAR strategies yielded nearly identical outcomes to the natural course, and this lack of effect was consistent across patient subgroups, including those with diabetes. Importantly, we did not observe signals of harm with the personalized strategy, in contrast to what was reported in the CONTROLING trial, particularly among non-diabetic and surgical patients. These results indicate that while personalization may not be harmful, it is unlikely to provide major survival benefits without more advanced implementation systems.
Our findings also have important methodological and clinical implications. From a methodological standpoint, this study demonstrates the feasibility and utility of target trial emulation to explore complex, time-varying interventions in the ICU, a setting where large-scale clinical trials are often infeasible. Clinically, the results reinforce that maintaining moderate glucose control remains a safe and pragmatic approach in most ICU patients. However, as new technologies for real-time glucose monitoring and algorithm-driven insulin titration become more widely available, future investigations should revisit personalization strategies with a focus on improving time-in-target range and minimizing glycaemic variability, rather than simply achieving lower mean glucose levels.
Our cohort had a short ICU length of stay, reflecting contemporary discharge practices and case-mix at our centre. This does not imply that the g-formula extrapolated effects far beyond observed data for most patients; rather, because ICU discharge was treated as a competing event, most patients contribute limited ICU person-time during which ICU mortality can occur. Consequently, the causal contrast is concentrated in the early ICU period when glucose control is applied and when ICU deaths occur. This pattern is consistent with major glycaemic control trials, where ICU length of stay is typically only a few days,(2,7) and therefore any causal effect of glycaemic strategies on ICU mortality would also be expected to emerge early rather than after prolonged ICU follow-up. Importantly, restricting analyses to patients who remain in ICU for a specific period would condition on a post-baseline variable affected by illness severity and treatment response and would not emulate a trial randomized at ICU admission, potentially introducing selection (collider) bias. We therefore focused on prespecified subgroup analyses based on baseline severity and clinical phenotypes associated with prolonged ICU stay.
Strengths and limitations
This study has several strengths. First, we used granular, high-frequency EMR data to model dynamic glucose trajectories and apply interventions in a realistic way. Second, the application of the parametric g-formula allowed us to account for time-varying confounding and simulate interventions that mimic those of a randomized trial. Third, our cohort was large, unselected, and representative of contemporary ICU practice, improving the external validity of the findings. Fourth, our protocol incorporated rigorous treatment separation between strategies and confirmed good adherence within each simulated intervention arm, enhancing internal validity. Fifth, our consistent results across clinically relevant subgroups, including patients with and without diabetes, improve confidence in the generalisability of the primary findings. From a causal inference perspective, the parametric g-formula requires adequate overlap in covariate distributions across treatment histories. In our dataset, the observed glucose ranges and covariate distributions were well represented across exposure patterns, supporting the plausibility of the simulated interventions and reducing concerns regarding positivity violations. Finally, the inclusion of sensitivity analyses comparing parametric and non-parametric estimates provided reassurance regarding model specification.
However, several limitations must be acknowledged. The study was conducted at a single centre, potentially limiting generalisability. Our cohort was restricted to patients with an available HbA1c measurement. Although HbA1c testing at our institution is supported by an automated protocol for patients aged ≥ 54 years, younger patients or those without recent HbA1c testing may have been underrepresented. Therefore, the findings may be most generalizable to ICU populations where HbA1c testing is routinely performed or to older critically ill populations. While target trial emulation improves upon traditional observational analyses, it cannot establish causality and remains hypothesis-generating. Furthermore, although our model enforced 100% adherence to the simulated strategies, unlike real-world practice, such adherence is rarely achievable clinically. The interventions were defined at the level of glucose trajectories rather than specific treatment actions such as insulin administration. Therefore, our analysis estimates the effect of maintaining glucose exposure within predefined target ranges, rather than the effect of specific therapeutic strategies used to achieve those targets. This trajectory-based intervention reflects a conceptual causal estimand representing sustained glycaemic control under full adherence but does not capture potential risks associated with treatment actions, including hypoglycaemia resulting from insulin administration. Because glucose values were constrained within the intervention range, the risk of hypoglycaemia induced by treatment could not be directly evaluated. As a result, our findings should be interpreted as the effect of glycaemic exposure patterns rather than the safety or efficacy of specific glucose-lowering treatments. This limitation is inherent to trajectory-based causal inference methods using observational data, where interventions are defined on exposure trajectories rather than clinical actions. The glucose control observed in the natural course differed meaningfully from the NICE-SUGAR target, suggesting variability in actual practice and possible contamination of "usual care". Missing time-varying covariates during follow-up were handled using last observation carried forward. While this approach reflects clinical practice, where treatment decisions are based on the most recent available measurement, it may introduce bias if measurement frequency is related to clinical deterioration or prognosis. However, in our single-centre ICU, laboratory and physiological measurements are obtained frequently, particularly during clinical instability, reducing the likelihood of prolonged reliance on carried-forward values. Nonetheless, residual bias related to informative measurement frequency cannot be excluded. Finally, we were not able to evaluate the effect of hypoglycaemia directly, which has been a key safety concern in prior trials and may modulate any benefit of tighter control.
CONCLUSION
In this large, single-centre target trial emulation using high-resolution intensive care unit data, we found no mortality benefit from personalizing glycaemic control based on admission HbA1c compared to standard glucose management strategies, including the NICE-SUGAR approach. The findings were consistent across all patient subgroups, including those with diabetes, and suggest that HbA1c-guided targets are unlikely to meaningfully impact 30-day resolution intensive care unit mortality under current implementation methods. While these results do not support a shift toward personalized glucose targets in routine ICU care, they highlight the importance of treatment adherence and the potential of advanced modelling techniques to evaluate complex interventions. As technologies for continuous glucose monitoring and closed-loop insulin delivery evolve, future studies should revisit the role of personalization with a focus on time-in-range, safety, and functional outcomes beyond mortality.
-
Key massage
Target trial emulation of 3,541 intensive care unit patients shows no mortality benefit from personalizing glucose targets using admission HbA1c. Individualized, NICE-SUGAR and usual-care strategies, all produced nearly identical 30-day intensive care unit mortality.
-
USE OF ARTIFICIAL INTELLIGENCE
The authors declare that no Artificial Intelligence tools were used in the preparation of this manuscript.
-
Publisher's note
AVAILABILITY OF DATA AND MATERIALS
The contents are already available
REFERENCES
- 1 Krinsley JS, Deane AM, Gunst J. The goal of personalized glucose control in the critically ill remains elusive. Intensive Care Med. 2021;47(11):1319-21.
- 2 NICE-SUGAR Study Investigators; Finfer S, Chittock DR, Su SY, Blair D, Foster D, Dhingra V, et al. Intensive versus conventional glucose control in critically ill patients. N Engl J Med. 2009;360(13):1283-97.
- 3 Honarmand K, Sirimaturos M, Hirshberg EL, Bircher NG, Agus MS, Carpenter DL, et al. Society of Critical Care Medicine Guidelines on Glycemic Control for Critically Ill Children and Adults 2024. Crit Care Med. 2024;52(4):e161-81.
- 4 Lovic D, Piperidou A, Zografou I, Grassos H, Pittaras A, Manolis A. The growing epidemic of diabetes mellitus. Curr Vasc Pharmacol. 2020;18(2):104-9.
- 5 Poole AP, Finnis ME, Anstey J, Bellomo R, Bihari S, Biradar V, et al.; LUCID Study Investigators and the Australian and New Zealand Intensive Care Society Clinical Trials Group (ANZICS CTG). The Effect of a liberal approach to glucose control in critically ill patients with type 2 diabetes: a multicenter, parallel-group, open-label randomized clinical trial. Am J Respir Crit Care Med. 2022;206(7):874-82.
- 6 Sherwani SI, Khan HA, Ekhzaimy A, Masood A, Sakharkar MK. Significance of HbA1c test in diagnosis and prognosis of diabetic patients. Biomark Insights. 2016;11:95-104.
- 7 Bohé J, Abidi H, Brunot V, Klich A, Klouche K, Sedillot N, et al.; CONTROLe INdividualisé de la Glycémie (CONTROLING) Study Group. Individualised versus conventional glucose control in critically ill patients: the CONTROLING study-a randomized clinical trial. Intensive Care Med. 2021;47(11):1271-83.
- 8 Lazar HL, McDonnell MM, Chipkin S, Fitzgerald C, Bliss C, Cabral H. Effects of aggressive versus moderate glycemic control on clinical outcomes in diabetic coronary artery bypass graft patients. Ann Surg. 2011;254(3):458-63.
- 9 Desai SP, Henry LL, Holmes SD, Hunt SL, Martin CT, Hebsur S, et al. Strict versus liberal target range for perioperative glucose in patients undergoing coronary artery bypass grafting: a prospective randomized controlled trial. J Thorac Cardiovasc Surg. 2012;143(2):318-25.
- 10 Gunst J, Debaveye Y, Güiza F, Dubois J, De Bruyn A, Dauwe D, et al.; TGC-Fast Collaborators. Tight blood-glucose control without early parenteral nutrition in the ICU. N Engl J Med. 2023;389(13):1180-90.
- 11 Krinsley JS, Rule P, Pappy L, Ahmed A, Huley-Rodrigues C, Prevedello D, et al. The interaction of acute and chronic glycemia on the relationship of hyperglycemia, hypoglycemia, and glucose variability to mortality in the critically ill. Crit Care Med. 2020;48(12):1744-51.
- 12 Fong KM, Au SY, Ng GW. Glycemic control in critically ill patients with or without diabetes. BMC Anesthesiol. 2022;22(1):227.
- 13 Hernán MA, Sauer BC, Hernández-Díaz S, Platt R, Shrier I. Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analyses. J Clin Epidemiol. 2016;79:70-5.
- 14 Hernán MA, Brumback B, Robins JM. Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV-positive men. Epidemiology. 2000;11(5):561-70.
- 15 Bolli GB, Porcellati F, Lucidi P, Fanelli CG. The physiological basis of insulin therapy in people with diabetes mellitus. Diabetes Res Clin Pract. 2021;175:108839.
- 16 Alhatemi G, Aldiwani H, Alhatemi R, Hussein M, Mahdai S, Seyoum B. Glycemic control in the critically ill: less is more. Cleve Clin J Med. 2022;89(4):191-9.
Edited by
-
Responsible editor:
Regis Goulart Rosa https://orcid.org/0000-0001-7881-9866
