Perioperative Glycemic Control and Outcomes in Cardiac Surgery: A Retrospective Cohort Study

Introduction

Perioperative hyperglycemia is a common metabolic response to surgical stress, particularly in cardiac surgery, and has been associated with increased morbidity and mortality [1–6]. It occurs in most patients with diabetes (approximately 60–90%) and in a substantial proportion of patients without a prior diagnosis of diabetes (around 60%) [7]. This transient but significant elevation in blood glucose levels results from reduced insulin secretion, increased insulin resistance, and the release of counterregulatory hormones, such as cortisol and catecholamines [1,4]. From a pathophysiological perspective, hyperglycemia impairs immune function, delays tissue repair, and reduces leukocyte activity, thereby increasing susceptibility to perioperative infection, including surgical site infections (SSIs) [4,9]. Persistent postoperative hyperglycemia has also been associated with morbidity, such as acute kidney injury and delirium, and increased mortality [3–8].

Despite its clinical relevance, optimal perioperative glycemic management remains controversial. Over the past 2 decades, clinical trials and perioperative cohort studies have reported conflicting findings regarding glycemic targets in surgical and critical care populations [10–13]. Early studies suggested that intensive insulin therapy could reduce morbidity, mortality, and hospital length of stay (LOS) [10,14]. However, subsequent large randomized controlled trials, most notably the NICE-SUGAR study, observed that strict glycemic targets (81–108 mg/dL) were associated with increased mortality, largely driven by a higher incidence of severe hypoglycemia [11]. Additional investigations have shown that intensive insulin therapy can increase the risk of hypoglycemia-related complications, including procoagulant effects, sympathetic activation, neurological injury, and autonomic dysfunction [15–18]. Collectively, these findings contributed to a shift away from strict glycemic control toward safer, moderate targets in perioperative and critical care settings [1,4,15,16].

Recent cardiac surgery-specific evidence further informs this debate. A 2026 systematic review and meta-analysis of randomized trials focused on tight glucose control regimens for preventing SSIs after cardiac surgery highlighted that potential reductions in infectious complications must be weighed against the risk of clinically meaningful hypoglycemia, reinforcing the need for pragmatic target ranges and context-specific implementation [4]. Complementing trial evidence, contemporary observational cohorts in cardiopulmonary bypass (CPB) populations show that postoperative hyperglycemia is common and clinically consequential. In a 2025 cohort, postoperative hyperglycemia after CPB was associated with worse outcomes, including higher rates of pulmonary infection and acute kidney injury, as well as longer ventilation time and prolonged ICU and hospital stays [5]. Similarly, analysis of the MIMIC-IV database (2024) found that severe hyperglycemia within the first 24 h of ICU admission after cardiac surgery was associated with increased risk of early acute kidney injury and higher ICU and hospital mortality [6]. These findings underscore that perioperative hyperglycemia is not only frequent but also is linked to multiple clinically relevant postoperative complications [4–6].

Hyperglycemia and hypoglycemia are now both recognized as independent risk factors for adverse outcomes in critically ill and surgical populations, with evidence supporting a U-shaped relationship between perioperative glucose levels and mortality [1,2,19]. Current guidelines, including those from the American Diabetes Association, the Centers for Disease Control and Prevention, and the Society of Thoracic Surgeons, recommend moderate perioperative glycemic control, generally maintaining glucose levels below 180 mg/dL and within a range of approximately 80 to 180 mg/dL [20–25]. Despite these recommendations, important uncertainties remain regarding optimal glycemic targets in cardiac surgery, particularly in real-world practice where adherence to guideline-based protocols is variable [1,4,10–14,20–25].

Furthermore, most available evidence originates from high-income countries and highly controlled clinical environments, which may limit external validity in resource-limited settings. Data from low- and middle-income healthcare systems, where perioperative management protocols, patient risk profiles, and infection burdens may differ, remain scarce. This is a critical knowledge gap, particularly regarding the impact of perioperative glycemic control on postoperative infection and mortality among cardiac surgery patients in these settings [1].

Therefore, this study primarily aimed to evaluate the relationship between perioperative glycemic control and postoperative infection and mortality among cardiac surgery patients in a tertiary cardiovascular center. Secondarily, the study aimed to assess compliance with the blood glucose target (BGT) of 70 to 180 mg/dL and the occurrence of hypoglycemia and hyperglycemia.

Material and Methods

STUDY DESIGN AND SETTING:

This retrospective cohort study was conducted at the Cardiology and Transplant Institute of the Federal District (ICDF), Brazil, a single tertiary cardiac center, between January 2021 and December 2022. The primary outcomes were in-hospital mortality and postoperative infection. Secondary outcomes included compliance with the BGT of 70 to 180 mg/dL, as well as occurrence of hypoglycemia (<70 mg/dL) and hyperglycemia (>180 mg/dL).

ICDF is a philanthropic, non-profit institution that provides high-complexity cardiovascular care and transplant services and operates as a complementary provider to the Federal District public health network. It serves as a referral center for patients from the Federal District and other Brazilian states, particularly in the North and Central-West regions, and most of its patients are treated through the Brazilian Unified Health System (SUS).

STUDY POPULATION AND DATA COLLECTION:

All consecutive adult patients (≥18 years) who underwent elective or emergency cardiac surgery and had a postoperative hospital stay of >72 h during the study period were eligible. Patients were excluded if their medical records were incomplete or if they had diabetic ketoacidosis or perioperative hypoglycemic coma. Incomplete records were defined as missing key variables required for exposure and/or outcome classification, and/or missing perioperative glucose measurements that precluded assessment of glycemic control.

Data were extracted from the ICDF electronic health record system. Perioperative capillary blood glucose was measured using calibrated point-of-care glucometers according to institutional routine. For this study, the perioperative period was defined as the interval from the beginning of anesthesia induction to hospital discharge or in-hospital death. Accordingly, all available glucose measurements recorded in the electronic health records within this time window were retrieved for analysis. This extended observation window was chosen to capture the full period of postoperative vulnerability, including early metabolic instability and later complications such as infection. Because glucose measurements were obtained as part of routine clinical care rather than a standardized research protocol, the number and timing of measurements varied across patients depending on clinical condition and monitoring intensity.

VARIABLES AND OUTCOMES:

Collected variables included demographic characteristics (age and sex), surgery type (myocardial revascularization, heart transplantation, heart valve, and other surgery), body mass index (BMI), and comorbidities (diabetes mellitus, systemic arterial hypertension, and dyslipidemia).

The BGT range of 70 to 180 mg/dL was selected in accordance with current guidelines for perioperative and critically ill patients, which recommend moderate glycemic control to minimize the risks of both hyperglycemia and hypoglycemia [19–25].

The primary outcomes were in-hospital mortality and postoperative infection. Secondary outcomes included compliance with the BGT of 70 to 180 mg/dL, as well as an episode of hypoglycemia (<70 mg/dL) and hyperglycemia (>180 mg/dL).

Compliance with blood glucose target (BGT) of 70 to 180 mg/dL was defined as maintaining glucose values within this range in ≥70% of all glucose measurements recorded during the perioperative period defined above. Hypoglycemia was defined as at least 1 glucose measurement (<70 mg/dL), and hyperglycemia as at least 1 glucose measurement (>180 mg/dL) during the same period. Because these definitions were based on threshold values, hypoglycemia and hyperglycemia were not mutually exclusive, and some patients could meet criteria for both conditions during the same hospitalization, reflecting glycemic instability.

Postoperative infection was defined pragmatically as documented in the medical record by the treating clinical team and associated with the initiation or escalation of systemic antibiotic therapy during the postoperative hospitalization period. When available, clinical, laboratory, and microbiological information recorded in the chart supported the diagnosis in accordance with routine institutional practice [26]. Because of the retrospective design, infections were not adjudicated using standardized surveillance criteria for all cases (such as CDC definitions for surgical site infection, pneumonia, or bloodstream infection). Therefore, outcome classification reflects routine clinical diagnosis and may include both microbiologically confirmed and clinically suspected infections [26–29].

STATISTICAL ANALYSIS:

Continuous variables were reported as means with standard deviations (SD) or as medians with interquartile ranges (IQR), depending on whether the Shapiro-Wilk test indicated normality. Categorical variables were summarized using absolute and relative frequencies. Missing data were handled using complete-case analysis without imputation. Overall, missingness was minimal: BMI was missing for 5 patients. No outliers were identified. Given the low proportion of missing data, no additional sensitivity analyses for missingness were performed.

In the univariate analysis, comparisons of factors associated with in-hospital mortality and postoperative infection were performed using the t test or the Mann-Whitney U test for continuous variables, as appropriate, and the Pearson chi-square test or Fisher test for categorical variables.

Separate multivariable logistic regression models were fitted for each primary outcome: 1 model for postoperative infection and 1 for in-hospital mortality. Variables with P values <0.05 in the univariate analyses, as well as clinically relevant variables with P<0.20, were considered for inclusion in the multivariable models using the Enter method. Hypoglycemia and hyperglycemia were treated as separate binary indicators of dysglycemia rather than mutually exclusive categories. Therefore, patients who met criteria for both conditions contributed to both variables in the analyses and were not excluded or reclassified. To reduce the risk of overfitting, candidate predictors were restricted to variables with stronger statistical and/or clinical relevance, and the final models were kept as parsimonious as possible. The final in-hospital mortality model included 7 predictors: BGT compliance, age, sex, systemic arterial hypertension, corticosteroid use, underweight, and heart valve/other surgery. Because only 14 in-hospital deaths were observed, this corresponded to an events-per-variable ratio of approximately 2.0. Therefore, although predictors were selected with emphasis on clinical plausibility and model parsimony, the mortality model should be interpreted cautiously as exploratory, given the potential for overfitting and unstable estimates. The final postoperative infection model included 4 predictors: hyperglycemia >180 mg/dL, corticosteroid use, underweight, and heart valve surgery. Diabetes mellitus was evaluated as a clinically relevant baseline covariate. However, it was not retained in the final multivariable models, in accordance with the prespecified variable-selection criteria and the need to preserve parsimony given the limited number of outcome events. In the postoperative infection model, hypoglycemia was considered a candidate predictor because of its clinical relevance and its association in univariate analysis. However, it was not retained in the final model to preserve parsimony, given the limited number of outcome events. Hyperglycemia showed a stronger univariate association and was therefore prioritized as the dysglycemia indicator in the final postoperative infection model. Model performance was assessed with the Hosmer-Lemeshow test and area under the ROC curve (AUC). Multicollinearity was assessed using tolerance values >0.10 and variance inflation factors (VIF) <10. Final results are presented as adjusted odds ratios (ORs) with corresponding 95% confidence intervals (CIs).

As a sensitivity analysis to mitigate small-sample bias and potential separation, we also fitted Firth’s penalized logistic regression models for the primary outcomes. In addition, we calculated E-values to quantify the minimum strength of association that an unmeasured confounder would need to have with both the exposure and the outcome to fully explain away the observed associations between (i) compliance with the blood glucose target (BGT) of 70 to 180 mg/dL and in-hospital mortality and (ii) hyperglycemia >180 mg/dL and postoperative infection. E-values were computed using an online E-value calculator (https://www.evalue-calculator.com/) [30,31].

Analyses were performed using R version 4.3.2 and Jamovi version 2.3.24. Statistical significance was defined as P≤0.05.

ETHICAL CONSIDERATIONS:

The research protocol was approved by the Research Ethics Committee of the Cardiology and Transplant Institute of the Federal District, Brazil, which waived informed consent due to the study’s retrospective nature. The study was conducted in accordance with the Declaration of Helsinki. All data were handled in compliance with applicable local data protection regulations. Patient information was de-identified and anonymized prior to analysis, and all records were stored securely to ensure confidentiality and data privacy.

Results

IN-HOSPITAL MORTALITY:

In the univariate analysis, compliance with the BGT within 70 to 180 mg/dL was associated with reduced in-hospital mortality (57.1% versus 80.4%, P=0.044). In comparison, corticosteroid use (78.6% versus 31.5%, P<0.001) and at least 1 episode of hypoglycemia <70 mg/dL (85.7% versus 26.6%, P<0.001) were significantly associated with increased in-hospital mortality. Systemic arterial hypertension was less frequent among non-survivors (35.7% vs 64.3%, p=0.035). Non-survivors were also older (64.1±11.0 versus 59.3±13.2 years) and more frequently female (57.1% versus 37.1%), although these differences did not reach statistical significance in univariate analysis (Table 2).

In the multivariate analysis, compliance with BGT within 70 to 180 mg/dL was independently associated with reduced in-hospital mortality (OR 0.10, 95% CI 0.02–0.59; P=0.011). Conversely, increasing age (per year, OR 1.21, 95% CI, 1.08–1.36; P=0.002), corticosteroid use (OR 13.21, 95% CI, 2.20–79.42; P=0.005), and female sex (OR 5.60, 95% CI, 1.09–29.10; P=0.039) were independently associated with increased in-hospital mortality. Systemic arterial hypertension was associated with lower in-hospital mortality (OR 0.08, 95% CI, 0.01–0.06; P=0.018), suggesting a possible “hypertension paradox” effect in this cohort (Table 3).

The regression model for in-hospital mortality demonstrated satisfactory calibration (Hosmer-Lemeshow P=0.74) and strong discrimination (AUC=0.88). In sensitivity analyses, the E-value of the adjusted OR for the association between BGT within 70 to 180 mg/dL and in-hospital mortality was 19.49, with a 95% CI of 2.78, indicating that only very strong unmeasured confounding could fully account for the observed association.

Firth’s penalized model yielded consistent directions of association, with BGT compliance associated with lower mortality (OR 0.23; 95% CI 0.06–0.88) and age (OR 1.11 per year; 95% CI 1.04–1.19), and corticosteroid therapy (OR 8.94; 95% CI 2.10–38.01) associated with higher mortality. The association with female sex was attenuated and did not reach statistical significance in this sensitivity analysis (OR 2.49; 95% CI 0.72–8.62; P=0.149).

POSTOPERATIVE INFECTION:

In the univariate analysis, patients who developed infection had significantly higher rates of corticosteroid use (53.2% versus 9.5%, P<0.001) and at least 1 episode of hyperglycemia >180 mg/dL (91.5% versus 69.8%, P<0.001) than those without infection. Myocardial revascularization procedures were more frequent among patients without infection, whereas valve and transplant surgeries predominated in the infected group (P<0.001) (Table 4).

In the multivariate analysis, both at least 1 episode of hyperglycemia >180 mg/dL (OR 3.77; 95% CI, 1.38–10.35; P=0.010) and corticosteroid therapy (OR 9.11; 95% CI, 3.38–24.06; P<0.001) remained independently associated with increased risk of postoperative infection (Table 5).

The regression model for postoperative infection demonstrated adequate calibration (Hosmer-Lemeshow P=0.62) and strong discrimination (AUC=0.82). In sensitivity analyses, the E-value of the adjusted OR for the association between hyperglycemia (>180 mg/dL) and postoperative infection was 3.77 (95% CI: 1.63).

Using Firth’s penalized logistic regression, hyperglycemia >180 mg/dL (OR 3.79; 95% CI 1.41–10.21) and corticosteroid therapy (OR 9.21; 95% CI 3.64–23.30) remained independently associated with postoperative infection.

Discussion

In this single-center Brazilian cohort study, perioperative dysglycemia was common and associated with adverse outcomes after cardiac surgery. Compliance with the BGT of 70 to 180 mg/dL was independently associated with lower in-hospital mortality. In contrast, the occurrence of at least one episode of hyperglycemia >180 mg/dL was independently associated with postoperative infection. These findings are biologically plausible as the stress response to major surgery activates inflammatory and neuroendocrine pathways that promote perioperative glucose fluctuations and may contribute to immune dysfunction and impaired microcirculatory perfusion [9,16–18].

Importantly, this study evaluated glycemic patterns observed during routine clinical care rather than the effect of a standardized therapeutic protocol. Therefore, the term perioperative glycemic control in this report refers to target-range compliance and dysglycemic exposure, rather than to protocolized insulin management strategies. In addition, because glucose measurements were analyzed throughout the hospitalization period, these findings should be interpreted as reflecting overall in-hospital glycemic patterns rather than glycemic management at a single postoperative time point. In this regard, our data support a pragmatic interpretation: maintaining glucose within a moderate target range may reflect a safer balance between avoiding sustained hyperglycemia and limiting iatrogenic hypoglycemia, which remains a key concern in perioperative and ICU settings [19–25].

Notably, the incidence of postoperative infection (59.9%) and in-hospital mortality (8.9%) in this Brazilian tertiary cohort appears higher than rates commonly reported in high-income settings [25]. This difference may reflect a combination of case-mix complexity (eg, urgent procedures, transplantation, higher comorbidity burden), structural constraints affecting ICU staffing, monitoring intensity, and protocol adherence, and differences in perioperative care resources. The relatively high postoperative infection rate should also be interpreted considering the pragmatic outcome definition adopted in this retrospective study [26]. Infections were identified based on documentation in the medical record by the treating team, together with initiation or escalation of systemic antibiotic therapy during hospitalization, rather than through formal adjudication using standardized surveillance criteria for specific categories such as surgical site infection, pneumonia, or bloodstream infection. Consequently, the outcome may have included both microbiologically confirmed and clinically suspected infections managed in routine practice, thereby increasing the sensitivity of case capture while limiting direct comparability with studies using narrower diagnostic definitions. In this context, perioperative glycemic control may represent both a potentially modifiable exposure and a marker of perioperative instability, underscoring the value of vigilant monitoring and protocolized management tailored to local capacity, alongside broader system-level strategies such as standardizing insulin pathways, staff training, and infection-prevention bundles [1,20].

The association between hyperglycemia (>180 mg/dL) and infection is biologically plausible and aligns with prior evidence showing that perioperative hyperglycemia can impair innate immune responses, reduce neutrophil function, and compromise wound healing, thereby increasing bacterial proliferation and susceptibility to infectious complications [4,9,12]. Hyperglycemia-induced oxidative stress and endothelial dysfunction can further compromise tissue perfusion and resistance to infection [30,31]. Contemporary reviews in cardiac surgery emphasize that stress hyperglycemia is common and clinically relevant, while also highlighting the ongoing controversy regarding the optimal BGT and the competing risks of hypoglycemia with tighter control [1,4]. A recent large systematic review and meta-analysis of randomized trials in cardiac surgery found that tight glucose control (upper target ≤150 mg/dL) was associated with a lower risk of surgical site infection but at the cost of a higher risk of hypoglycemia, with no significant effect on short-term mortality; ICU stay was modestly shorter [4]. These data help contextualize our results: rather than supporting very tight targets, our findings reinforce the rationale for moderate perioperative targets (70–180 mg/dL), aiming to reduce infectious risk while limiting hypoglycemia exposure [19–25].

Recent observational studies further support the clinical burden and correlates of postoperative hyperglycemia [5,6]. In a cohort of 1008 patients undergoing cardiopulmonary bypass surgery, postoperative hyperglycemia (≥180 mg/dL on 2 occasions within 24 h) occurred in approximately 65% of patients. It was associated with higher rates of acute kidney injury, delirium, pulmonary infection, longer days in mechanical ventilation, and longer ICU and hospital LOS. Risk factors included age, female sex, diabetes, pulmonary infection, cross-clamp time, and intraoperative peak glucose [5]. Beyond infection, early hyperglycemia has been linked to other major postoperative complications. In a large MIMIC-IV analysis of post-cardiac surgery ICU admissions, severe admission hyperglycemia (≥200 mg/dL) within 24 h was associated with a higher incidence of 7-day acute kidney injury and higher ICU and hospital mortality, even after multivariable adjustment [6]. The high frequency of hyperglycemia in that cohort mirrors the substantial prevalence observed in our sample and supports interpreting perioperative hyperglycemia as a clinically meaningful marker associated with postoperative morbidity.

Our findings also are consistent with modern ICU trial evidence cautioning against causal interpretations that “tighter is better” across settings. In a large, randomized trial in critically ill patients not receiving early parenteral nutrition, targeting 80 to 110 mg/dL achieved lower glucose values but did not reduce ICU length of stay or 90-day mortality, with small absolute differences in severe hypoglycemia and largely similar secondary outcomes [13]. These results reinforce a balanced approach that prioritizes avoidance of sustained hyperglycemia while minimizing iatrogenic hypoglycemia, particularly in heterogeneous perioperative and ICU environments.

Corticosteroid therapy emerged as one of the strongest independent factors associated with both postoperative infection and in-hospital mortality. This finding aligns with prior evidence that steroid-induced hyperglycemia amplifies postoperative immune suppression and inflammatory dysregulation, creating a synergistic effect that heightens vulnerability to infection and death [32,33]. However, corticosteroid use may also reflect confounding by indication, as patients receiving steroids may have greater baseline severity or specific clinical contexts (eg, transplantation and immunosuppression). Beyond absolute glucose thresholds, the literature increasingly highlights glycemic variability as a prognostic marker. In a large meta-analysis of observational cohorts (n>16 000), higher acute glycemic variability was associated with increased in-hospital major adverse events and higher in-hospital mortality after cardiac surgery [36]. Although our study operationalized glycemic exposure using pragmatic definitions (BGT compliance and the presence of hypo-/hyperglycemia episodes), future studies should incorporate standardized variability metrics (eg, coefficient of variation) to test whether variability adds incremental prognostic value beyond threshold-based definitions.

Nonetheless, this study has limitations. First, its retrospective observational design precludes causal inference, and the observed associations may be influenced by residual or unmeasured confounding, including diabetes status and severity, obesity, baseline clinical risk, surgical complexity, cardiopulmonary bypass time, perioperative insulin management, and differential exposure to infection. Second, the single-center setting in Brazil and the relatively small sample size, particularly the limited number of mortality events, may restrict statistical power, increase the risk of model overfitting, and limit generalizability to other healthcare systems and populations. In addition, the multivariable mortality model included more predictors than would be supported by conventional events-per-variable recommendations, increasing the risk of overfitting and unstable estimates. Therefore, adjusted mortality estimates should be interpreted with caution and regarded as exploratory. To mitigate small-sample bias and potential separation, we performed sensitivity analyses using Firth’s penalized logistic regression. The direction and magnitude of the main associations were generally consistent, supporting model stability under penalization. Third, the retrospective design and exclusion of patients with incomplete records may have introduced selection bias if missing data were not random. Fourth, glucose measurements were obtained as part of routine clinical care rather than according to a standardized measurement schedule. Therefore, variability in measurement method (capillary versus arterial or venous samples) and in the number and timing of assessments may have affected the classification of glycemic exposure and BGT compliance. Fifth, postoperative infection was defined using a pragmatic clinical approach based on chart documentation and antibiotic therapy rather than standardized surveillance definitions, which may have increased the reported incidence. Finally, the single-center design may limit generalizability to other healthcare settings.

Despite these constraints, the study provides a clinically relevant description of perioperative glycemic patterns and their associations with postoperative infection and in-hospital mortality in a real-world tertiary cardiac surgery setting. To further evaluate robustness to unmeasured confounding, we calculated E-values for the primary associations, which suggest that unmeasured confounders were unlikely to explain the entirety of the effect observed in our study, to the point of modifying the observed results [34,35]. Given these methodological constraints, the findings should be interpreted cautiously and should be considered hypothesis-generating rather than practice-changing. Prospective multicenter studies, ideally with standardized insulin protocols and measurement schedules and external validation, are needed to confirm these associations and to inform perioperative glucose management strategies across diverse healthcare settings. Finally, the models showed good internal discrimination (AUC 0.82 for infection and 0.88 for mortality), suggesting reasonable separation of outcome groups within this cohort. However, performance estimates require confirmation in independent prospective datasets before clinical risk stratification applications. From a practical standpoint, our results support implementing standardized glycemic control protocols, preferably nurse-driven, protocolized, and supported by timely glucose monitoring, to enhance safety and consistency across teams. Close coordination among intensivists, anesthesiologists, and nursing staff is essential to reduce glycemic excursions and optimize outcomes, and even in resource-limited cardiac ICUs, protocol-driven approaches may help reduce infections and LOS [37,38].

Conclusions

This single-center retrospective study suggests that perioperative glycemic patterns are associated with clinically relevant outcomes after cardiac surgery. Episodes of hyperglycemia >180 mg/dL were independently associated with postoperative infection, whereas compliance with a moderate BGT (70–180 mg/dL) was independently associated with lower in-hospital mortality. Because residual confounding, limited event counts, and lack of external validation may influence these estimates, the results should be considered hypothesis-generating. They should be confirmed in multicenter prospective cohorts and randomized studies before being used to guide guideline-level changes. Future research should integrate standardized glycemic variability indices (such as the coefficient of variation) to evaluate whether variability contributes additional prognostic value beyond threshold-based classifications.