Critical Illness-Associated Hyperglycemia and New-Onset Diabetes: A Retrospective Cohort Study

Abstract

Objective:

To evaluate the relationship between critical illness-associated hyperglycemia (CIAH) during ICU treatment and the development of incident diabetes in ICU survivors without pre-existing diabetes or prediabetes.

Design:

Retrospective observational study.

Setting:

Four university hospital ICUs in Stockholm, Sweden.

Patients:

A total of 6633 ICU survivors admitted between 2010 and 2021, with no prior diabetes diagnosis recorded in the Swedish National Diabetes Register (NDR) and a glycated hemoglobin A1c level below 42 mmol/mol (6%) at admission.

Interventions:

None.

Measurements and Main Results:

CIAH was defined as insulin administration to maintain blood glucose between 6 and 10 mmol/L (108–180 mg/dL) in ICU. Incident diabetes was defined as an NDR registration after ICU discharge, occurring beyond 30 days until September 2023. Overall, 3100 (46.7%) patients developed CIAH in the ICU. The 5-year cumulative diabetes incidence was higher in patients with CIAH (4.1%, 95% CI, 3.4–4.9%) compared with those without CIAH (1.8%, 95% CI, 1.3–2.3%). On multivariable Cox regression, the adjusted hazard ratio for incident diabetes was 2.15 (95% CI, 1.52–3.03) in patients with CIAH. Similarly, multivariable competing risk analysis revealed an adjusted sub-hazard ratio of 2.20 (95% CI, 1.57–3.08) for CIAH.

Conclusions:

CIAH in ICU patients without pre-existing diabetes or prediabetes was associated with a higher risk of developing incident diabetes within 5 years of ICU discharge.

KEY POINTS.

Question: Does critical illness-associated hyperglycemia (CIAH) increase the risk of developing diabetes in ICU survivors without pre-existing diabetes or prediabetes?

Findings: In a retrospective cohort study of 6633 ICU survivors, those with CIAH, defined as requiring insulin administration to maintain blood glucose between 6 and 10 mmol/L (108–180 mg/dL), had a significantly higher 5-year cumulative incidence of diabetes (4.1%) compared with those without CIAH (1.8%). These results were statistically significant.

Meaning: CIAH is associated with an increased risk of developing diabetes after ICU discharge, suggesting potential value of proactive diabetes screening and structured follow-up care for ICU survivors who require insulin during critical illness.

Critical illness often causes acute metabolic disturbances, including insulin resistance, stress hyperglycemia (SH), and the need for exogenous insulin to achieve glucose control. Studies on ICUpatients without known diabetes have shown that SH, commonly defined as blood glucose levels greater than 7.8–11.1 mmol/L (140–200 mg/dL) during the first 24 hours of ICU admission, is associated with a two-fold increase in the risk of developing diabetes later in life (1, 2). However, these findings may overestimate the risk due to potential inclusion of patients with undiagnosed prediabetes or diabetes, a subgroup that may account for approximately 15% of ICU patients (3).

In addition, clinical practices for glucose management in ICUs might influence the observed incidence of SH. For example, insulin therapy is typically initiated when blood glucose levels reach 10 mmol/L (180 mg/dL), which prevents some patients from exceeding the 11.1 mmol/L (200 mg/dL) threshold used in previous studies, potentially underestimating SH prevalence. Furthermore, hyperglycemia often appears only after adequate nutritional support, which may occur beyond the first 24 hours of ICU admission (4). Thus, defining SH based on glucose measurements within the initial 24-hour period may miss the progression of later hyperglycemia and insulin resistance.

To address these limitations, we evaluated critical illness-associated hyperglycemia (CIAH), defined as the requirement for insulin therapy at any point during the ICU stay, and the rate of new-onset diabetes among ICU survivors without a history of diabetes or evidence of prediabetes at admission. We hypothesized that CIAH, as an indicator of the persistence and severity of metabolic disturbances during critical illness, would be independently associated with an increased rate of developing new-onset diabetes.

METHODS

The study was approved by the Swedish Ethical Review Authority on June 16, 2021 (approval number 2021-02699, study title: “Blood glucose dysregulation in ICU patients—risk factors and outcome”) with a waiver of informed consent. The study was conducted in accordance with the ethical principles of the Declaration of Helsinki (1975 and its later amendments) and the ethical standards of the responsible national and institutional committees on human research.

Study Design and Patient Selection

We conducted a retrospective cohort study of adult (18 yr old or older) patients admitted to four ICUs at the Karolinska University Hospitals (Solna site and Huddinge site) in Stockholm, Sweden between January 2010 and June 2021. These ICUs included two general ICUs (one at each site), a neurointensive care unit, and a cardiothoracic ICU (both at Solna site). We excluded patients without a valid social security number, patients without any recorded data in the local ICU patient data management (PDMS) system, patients with pre-existing prediabetes or diabetes on ICU admission, and patients who died in the ICU, or within thirty days of ICU discharge. For patients with multiple ICU admissions, only the first episode was considered.

Transfers to different ICUs in the same network and readmissions within 24 hours were considered one episode. Pre-existing diabetes was defined as a diabetes diagnosis recorded in the Swedish National Diabetes Register (NDR) either before ICU admission or within thirty days after ICU discharge. Prediabetes was defined as an glycated hemoglobin A1c (HbA1c) of 42–47 mmol/mol (6–6.5%) (International Expert Committee guidelines [5]) and undiagnosed diabetes as HbA1c greater than or equal to 48 mmol/mol (6.5%). HbA1c values were obtained from patients’ medical records if a sample was taken between 90 days before ICU admission and until 2 days after ICU admission. HbA1c was measured from whole blood using capillary electrophoresis with an International Federation of Clinical Chemistry and Laboratory Medicine calibrated testing method. At all study sites, insulin was administered by subcutaneous injections or IV infusion to maintain a blood glucose concentration of 6–10 mmol/L 108–180 mg/dL). Blood glucose was measured in arterial blood using the ABL800 Flex blood gas analyzer (Radiometer Medical A/S, Brønshøj, Denmark). HbA1c (expressed in mmol/mol) was measured in whole blood using the VARIANT II TURBO Hemoglobin Testing System analyzer (Bio-Rad Laboratories GmbH, Feldkirchen, Upper Bavaria, Germany).

Data Collection

We collected data from the electronic ICU PDMS Centricity Critical Care (GE Healthcare, Chicago, IL), from the hospital electronic health record system Take Care (CompuGroup Medical, Koblenz, Germany), and from the NDR. PDMS provided demographic data, ICU admission diagnoses, blood glucose levels, and information on subcutaneous and IV insulin administration and corticosteroid administration. Take Care provided demographic data, comorbidity data (International Classification of Diseases, 10th Revision codes) and information on date of death. From the NDR, which covers approximately 90% of all adults with diabetes in Sweden, we collected diabetes debut date, until September 15, 2023 which was end of follow-up for the study.

Data Processing and Validation

Our database integrates data from multiple sources, including automatically recorded data, manually entered curated data (e.g., from NDR), and manually recorded observations (e.g., direct drug administration, weight, and height). Manual entries are error prone due to human factors. During exploratory analysis, extreme outliers (e.g., height < 90 cm, weight > 300 kg, body mass index [BMI] > 90, or abnormally high insulin dosages) were primarily identified in manually entered data. To address this, we applied rigorous data validation procedures informed by clinical expertise, including manual review of patient records and cross-verification across data sources, before selecting values. This process reinforced the retention of physiologically plausible and clinically accurate data.

Exposure Variables

Primary exposure was CIAH, defined as requiring administration of insulin (any dose) to maintain a blood glucose between 6 and 10 mmol/L (108–180 mg/dL) while in the ICU (exposure period). Confounding variables included: age category (18–34 yr, 35–49 yr, 50–64 yr, 65 yr or older), sex, BMI category according to World Health Organization cutoffs (6) (normal [reference] range [18.5–24.9 kg/m²], underweight [< 18.5 kg/m²], overweight [25.0–29.9 kg/m²], obese [≥ 30 kg/m²], unknown BMI), HbA1c category according to the BiomarCaRE consortium (7) (< 34.4 mmol/mol [5.3%], 34.4–38.8 mmol/mol [5.3–5.7%], 38.8–41.9 mmol/mol [5.7–6%]), presence of pre-existing liver disease (yes/no), pre-existing renal disease (yes/no), cumulative hydrocortisone-equivalent dose (continuous), admission year (continuous) and variables reflecting illness severity (length of stay [continuous], mechanical ventilation [yes/no], renal replacement therapy [yes/no], and noradrenaline infusion rate > 0.1 µg/kg/min ≥10 min [yes/no]).

Outcomes

Primary outcome was time to new-onset diabetes from 30 days after ICU discharge as recorded in the NDR.

Statistical Analysis

Categorical data are presented as numbers (%) and continuous data are presented as medians with interquartile ranges (IQRs). Time to new-onset diabetes was estimated by cumulative incidence with Gray’s test. Hazards of new-onset diabetes was modeled by Cox regression (unadjusted; age-/sex-adjusted; fully adjusted). The fully adjusted model included: age, sex, BMI, cumulative hydrocortisone-equivalent dose in the ICU, length of stay in the ICU, admission year, pre-existing renal disease, pre-existing liver disease, invasive mechanical ventilation, renal replacement therapy, continuous noradrenaline infusion greater than 0.1 µg/kg/min for greater than or equal to 10 minutes, and HbA1c subcategory. Proportional hazards assumption was checked via log-log plots. Death was considered in a competing risk analysis using the Fine and Gray subdistribution hazard model. Time to death was presented using Kaplan-Meier curves with comparisons via log-rank test. Follow-up time for the survival analysis models commenced 30 days after ICU discharge until diabetes diagnosis, death, or censoring (NDR extraction date or maximum of 5 yr). Three sensitivity analyses assessed the robustness of our findings: 1) an analysis incorporating patients excluded due to missing HbA1c values. Missing HbA1c (yes/no) was included as a covariate in this analysis, 2) an analysis restricted to patients admitted from 2016 and onward, when routine HbA1c sampling was implemented. Routine HbA1c was defined as samples collected at admission day or the following 2 days (general and neuro ICUs); or either collected preoperatively or up to 2 days after admission (cardiothoracic ICU), and 3) an analysis adding the daily parenteral caloric intake as a predictor in the fully adjusted model. All analyses were two-sided (α < 0.05). Data analysis was performed using R Software (Version 4.4.0; R Foundation for Statistical Computing, Vienna, Austria).

RESULTS

Study Population

Patient selection is shown in Figure 1. 27,454 adult patients were admitted to our ICUs between January 1, 2010, and June 1, 2021. We excluded 341 patients without PDMS data, 5,611 with pre-existing diabetes, 3,057 who died in the ICU or within 30 days after ICU discharge, 10,588 with missing HbA1c data, and 1,224 with admission HbA1c greater than or equal to 42 mmol/mol (6%). Patients with and without available HbA1c showed statistically significant differences in some characteristics (Supplemental Table 1, https://links.lww.com/CCM/H803). The final study cohort consisted of 6,633 ICU patients with HbA1c less than 42 mmol/mol (6%) measured at a median (IQR) of 0.8 hours (0.3–1.7) before ICU admission.

Figure 1.

Patient selection. HbA1c= glycated hemoglobin A1c, PDMS = patient data management system.

Overall, 3100 (46.7%) patients developed CIAH. Insulin therapy was initiated after a median (IQR) of 6 hours (3–13) after ICU arrival. Median (IQR) blood glucose level before insulin initiation was 9.7 (8.5–11.1) mmol/L (175; 153–200 mg/dL). Compared with patients without CIAH, patients with CIAH were older, more often female, more often obese, had higher HbA1c, more comorbidities and differed in admission source (Table 1; and Supplemental Table 2, https://links.lww.com/CCM/H803). A detailed list of admission diagnoses is provided in Supplemental Table 3 (https://links.lww.com/CCM/H803). Patients with CIAH stayed longer in the ICU, and were more likely to receive invasive mechanical ventilation, renal replacement therapy, a noradrenaline infusion rate greater than 0.1 µg/kg/min for greater than or equal to 10 minutes, higher cumulative hydrocortisone-equivalent doses, and greater exposure to parenteral calories (Table 1). The availability of routinely collected HbA1c remained fairly constant from 2016 and onward across all ICUs (Supplemental Fig. 1, https://links.lww.com/CCM/H803).

TABLE 1.

Baseline Characteristics and Process of Care in the ICU

Characteristic	Critical Illness-Associated Hyperglycemia
	No, n = 3,533a	Yes, n = 3,100a	p b
Age, yr	55 (39, 68)	63 (51, 71)	< 0.001
Female sex	1,166 (33.0%)	1,145 (36.9%)	< 0.001
Body mass index, kg/m2c	25.1 (22.6, 28.1)	25.6 (23.0, 28.8)	< 0.001
HbA1c, mmol/mol	35.0 (33.0, 38.0)	36.0 (33.0, 39.0)	< 0.001
HbA1c, %	5.4 (5.2, 5.6)	5.4 (5.2, 5.7)	< 0.001
Comorbidity
Previous myocardial infarction	295 (8.3%)	296 (9.5%)	0.087
Congestive heart failure	243 (6.9%)	325 (10.5%)	< 0.001
Peripheral vascular disease	462 (13.1%)	565 (18.2%)	< 0.001
Cerebrovascular disease	530 (15.0%)	588 (19.0%)	< 0.001
Chronic obstructive pulmonary disease	175 (5.0%)	161 (5.2%)	0.7
Chronic other pulmonary disease	245 (6.9%)	222 (7.2%)	0.7
Rheumatic disease	150 (4.2%)	165 (5.3%)	0.040
Renal disease	173 (4.9%)	230 (7.4%)	< 0.001
Liver disease	170 (4.8%)	294 (9.5%)	< 0.001
Malignancy	543 (15.4%)	537 (17.3%)	0.032
Process of care
ICU length of stay, d	0.9 (0.7, 2.0)	2.0 (0.9, 6.5)	< 0.001
Invasive mechanical ventilation	2,413 (68.3%)	2,724 (87.9%)	< 0.001
Renal replacement therapy	104 (2.9%)	221 (7.1%)	< 0.001
Noradrenaline > 0.1 µg/kg/min ≥ 10 min	616 (17.4%)	1,410 (45.5%)	< 0.001
Corticosteroid treatment	518 (14.7%)	1,130 (36.5%)	< 0.001
Cumulative hydrocortisone-equivalent dose exposure in the ICU, among those who received steroids, mg	214 (100, 450)	700 (214, 1,440)	< 0.001
Mean daily parenteral calories, kcald	159 (101, 292)	224 (126, 407)	< 0.001

HbA1c = glycated hemoglobin A1c.

^aValues are n (%) or median (interquartile range).

^bWilcoxon rank sum test; Pearson’s χ².

^cData available for 3245 without critical illness-associated hyperglycemia and for 3021 patients with critical illness-associated hyperglycemia.

^dData available for 3375 without critical illness-associated hyperglycemia and for 3095 patients with critical illness-associated hyperglycemia.

Primary Outcome

Median (IQR) follow-up time was 64 months (39–87) for patients with CIAH and 64 months (39–82) for those without CIAH (p = 0.12). During follow-up, 230 patients developed new-onset diabetes with a 5-year cumulative incidence of 4.1% (95% CI, 3.4–4.9%) in the CIAH group and 1.8% (95% CI, 1.3–2.3%) in the no CIAH group (p < 0.001) (Fig. 2A). No statistically significant difference in 5-year mortality between the two groups was observed (p = 0.09) (Fig. 2B). In fully adjusted Cox regression, CIAH remained strongly associated with new-onset 5-year diabetes rate (adjusted HR 2.15, 95% CI, 1.52–3.03; p < 0.001). In addition, age between 50 years old or older and 64 years, BMI greater than or equal to 25 kg/m², and HbA1c greater than 38.8 mmol/mol (5.7%) were significantly associated with increased hazard of new-onset diabetes (Fig. 3). In the fully adjusted Fine and Gray model accounting for death as competing risk the subdistribution hazard ratio (sHR) for CIAH was 2.20 (95% CI, 1.57–3.08) (Table 2).

Figure 2.

Five year outcomes amont patients with and without critical illness-associated hyperglycemia (CIAH). A, Cumulative incidence curves demonstrating a higher indcidence of diabetes in patients with CIAH compared to those without (p < 0.001). B, Kaplan-Meier survival curves where no statistically significant difference in survival was observed between the groups (p = 0.09). The shaded regions represent 95% CIs.

Figure 3.

Adjusted hazard ratio (95% CI) for new-onset diabetes in the fully adjusted Cox regression model. The model was adjusted for age, sex, cumulative hydrocortisone-equivalent dose in the ICU, length of stay in the ICU, body mass index category, pre-existing renal or liver disease, admission year, invasive mechanical ventilation, renal replacement therapy, glycated hemoglobin A1c (HbA1c)-subcategory and noradrenaline treatment. CIAH = critical illness-associated hyperglycemia.

TABLE 2.

Cox Proportional Hazards Models and Fine and Gray Subdistribution Hazard Models for the 5-Year Rate of Diabetes Diagnosis in Patients With Critical Illness-Associated Hyperglycemia vs. Those Without

Model	Crude Estimate		Adjusted for Age and Sex		Fully Adjusteda,b
Cox Proportional Hazards Model	HR (95% CI)	p	HR (95% CI)	p	HR (95% CI)	p
No CIAH	1.00		1.00		1.00
CIAH	2.52 (1.83–3.48)	< 0.001	2.23 (1.61–3.08)	< 0.001	2.15 (1.52–3.03)	< 0.001

Fine and Gray Model	sHR (95% CI)	p	sHR (95% CI)	p	sHR (95% CI)	p
No CIAH	1.00		1.00		1.00
CIAH	2.50 (1.81–3.45)	< 0.001	2.26 (1.64–3.11)	< 0.001	2.20 (1.57–3.08)	< 0.001

CIAH = critical illness-associated hyperglycemia, HR = hazard ratio, sHR = subdistribution hazard ratio.

^aFully adjusted model includes age, sex, cumulative hydrocortisone-equivalent dose in the ICU, length of stay in the ICU, admission year, body mass index category, pre-existing renal or liver disease, invasive mechanical ventilation, renal replacement therapy, glycated hemoglobin A1c-subcategory and noradrenaline treatment.

^bPreexisting renal and liver disease were identified using International Classification of Diseases, 10th Revision codes recorded before ICU admission in the electronic health records. Liver disease included chronic liver disorders and complications such as varices or ascites; renal disease included chronic kidney disease and dialysis-related diagnoses.

The top rows present HR estimated using the Cox proportional hazards models under crude, age- and sex-adjusted, and fully adjusted conditions. The bottom rows present sHR estimated using the Fine and Gray model, accounting for competing risk of death, under the same levels of adjustment.

In sensitivity analyses (Supplemental Table 4, https://links.lww.com/CCM/H803) reintegrating the 10,533 patients with missing HbA1c and including missing-HbA1c as a covariate instead of HbA1c subcategories, CIAH remained strongly associated with 5-year diabetes risk (adjusted HR 2.35; 95% CI, 1.99–2.79, p < 0.001; adjusted sHR 2.37, 95% CI, 2.00–2.80). Missing‑HbA1c as covariate was not significant (HR 1.00, 95% CI, 0.84–1.19). When the cohort was restricted to the 4,915 patients admitted from 2016 onward during routine HbA1c sampling, the associations persisted (adjusted HR 2.30, 95% CI, 1.52–3.48, p < 0.001; adjusted sHR 2.37, 95% CI, 1.58–3.54, p < 0.001). In addition, adding daily parenteral caloric intake (available for 97.5% of patients) to the fully adjusted main model, caloric exposure had no independent effect on diabetes risk (HR per 100 kcal 0.99, 95% CI, 0.91–1.07, p = 0.80), whereas CIAH remained a significant predictor (adjusted HR 2.18, 95% CI, 1.54–3.10, p < 0.001). Finally, in the subgroup of 421 patients with sepsis (Supplemental Table 3, https://links.lww.com/CCM/H803), five patients with CIAH and three patients without developed new-onset diabetes at 5 years. Further subgroup analysis was not performed due to limited statistical power.

DISCUSSION

Key Findings

Among 6,633 ICU survivors without prediabetes or diabetes, 4.1% of patients with CIAH developed diabetes within 5 years vs. 1.8% without CIAH. After adjustment, CIAH remained associated with a two-fold higher hazard of new-onset diabetes, even when accounting for death as competing risk.

Relationship With Previous Studies

The relationship between SH or CIAH and post-ICU diabetes has been addressed in relatively few studies. An Australian registry-based study of 17,000 ICU survivors without diabetes (1) found SH (blood glucose ≥ 11.1 mmol/L [200 mg/dL] in the first 24 hr) in 17% of patients. Five years after hospital discharge, the cumulative diabetes incidence, was approximately 7% among patients with SH compared with 3.5% among those without SH. SH was independently associated with a two-fold higher hazard of new-onset diabetes. Similarly, a Scottish Registry-based study of over 1,800 ICU survivors without diabetes (8) reported SH (admission blood glucose > 11.1 mmol/L [200 mg/dL]) in 11% of patients. Three years after discharge, type 2 diabetes incidence was approximately 5% in patients with SH compared with 2.5% in those without. Admission blood glucose levels of 7–11.1 mmol/L (126–200 mg/dL) and greater than 11.1 mmol/L (200 mg/dL) were associated with 1.6-fold and 3.4-fold higher hazards, respectively, of developing diabetes, compared with blood glucose less than or equal to 7 mmol/L (126 mg/dL).

Unlike these studies focusing on early hyperglycemia, our study used a broader definition of hyperglycemia, encompassing arterial blood glucose levels and insulin requirements throughout the ICU stay. This likely explains the higher prevalence of hyperglycemia (47%) observed. Insulin therapy was initiated within 24 hours of ICU admission for most patients requiring insulin, at a median (IQR) blood glucose level of 9.7 (8.5–11.1) mmol/L (175; 153–200 mg/dL). This implies that only 25% of patients with CIAH and 12% of all patients met the SH threshold used in the Australian and Scottish studies. Our lower cumulative diabetes incidence likely reflects population differences and the lack of baseline HbA1c data in earlier studies, which may have included undiagnosed (pre)diabetes; nonetheless, the HRs for SH reported in those studies closely mirror our findings.

Several smaller, single-center studies used HbA1c, oral glucose tolerance tests (OGTT), and fasting glucose levels to assess prediabetes and diabetes. One study of 258 ICU patients without pre-existing diabetes or elevated fasting glucose/glucose intolerance found SH (at least two blood glucose readings greater than or equal to 7.8 mmol/L (140 mg/dL) during the ICU stay) in 35% of patients (9). The 5-year cumulative diabetes incidence was higher among patients with SH (16%) compared with those without SH (4%). Lack of HbA1c measurements at admission may underestimate pre-existing diabetes. Although the authors excluded undiagnosed diabetes using OGTT at discharge, this method could underestimate baseline diabetes due to impaired gastric emptying commonly associated with critical illness (10). Since OGTT results are depends on gastric emptying, this could affect the accuracy of their findings (11). These methodological differences are likely to contribute to the higher reported diabetes incidence compared with our findings.

Another study of 338 non-diabetic ICU patients with a length of stay greater than 48hours reported disturbed glucose metabolism (impaired fasting glucose or pathologic OGTT) in 38% of patients with SH (defined as blood glucose > 7.7 mmol/L [139 mg/dL] during the ICU stay) compared with 28% in normoglycemic patients (p = 0.065), and diabetes (defined by pathologic fasting glucose, OGTT or HbA1c approximately 8 mo after ICU discharge) in 9% vs. 4%, respectively (p = 0.246) (12). Furthermore, a study of 40 patients with SH (blood glucose > 11.1 mmol/L (200 mg/dL) in the first 24 hours or insulin treatment) and admission HbA1c less than 48 mmol/mol (6.5%) found that 43% had prediabetes and 37% fulfilled diabetes criteria at 3 months (13). The high diabetes incidence in these studies may stem from short-term follow-up periods (3–8 mo) capturing transient hyperglycemia, rather than stable diabetes diagnoses.

In addition to SH, one of the studies identified an unadjusted association between higher HbA1c levels, even when below 48 mmol/mol (6.5%), elevated BMI (a component of the Finnish diabetes risk score), and increased diabetes risk (12). Furthermore, ICU patients aged 50–69 years were more likely to develop diabetes than other age groups (1, 8). Similarly, our study demonstrates a clear, independent, and gradual increase in the hazard of new-onset diabetes with rising BMI above 25 kg/m² and increasing HbA1c within the normal range. In addition, we found a significant elevated hazard among patients aged 50–64years, reinforcing the importance of these risk factors in post-ICU diabetes development.

Implications of Study Findings

The association between CIAH and post-ICU diabetes has several clinical implications. CIAH may serve as a more comprehensive marker than early SH for identifying survivors at high risk, suggesting that those treated with insulin may benefit from structured diabetes screening and follow‑up as part of post‑intensive care syndrome management (14). The interplay between SH/CIAH, insulin resistance, and subsequent diabetes merits further investigation, since SH/CIAH may act as a trigger for diabetes through mechanisms such as β-cell dysfunction or persistent insulin resistance. Although our study does not establish a causal link, it underscores the need to determine whether improved glucose control in the ICU could reduce future diabetes risk. It is equally plausible that pre-existing metabolic dysfunction contributes to the observed dysglycemia. Early insulin resistance, characterized by reduced responsiveness in insulin-sensitive tissues, is often compensated by pancreatic β-cell hypersecretion, maintaining normoglycemia and normal HbA1c levels (15). Overt hyperglycemia and elevated HbA1c emerge only when β-cell compensation fails (16). Consequently, patients with early-stage insulin resistance may not exhibit hyperglycemia or elevated HbA1c prior to ICU admission.

Without pre-ICU assessments such as OGTT or C-peptide levels, it remains challenging to exclude prior metabolic disturbances. Although we adjusted for both HbA1c and BMI, these markers may not fully capture pre-existing insulin resistance, which may still occur in patients with normal BMI. This underscores the need for future studies to evaluate metabolic status more comprehensively before and after critical illness and to explore individualized glycemic management targets during ICU care.

Strengths and Limitations

Our study’s strengths include a large cohort with comprehensive electronic health records linked to the NDR, facilitating robust outcome assessment and adjustment for confounders. We employed a rigorous patient selection process, leveraging HbA1c values to minimize inclusion of individuals with pre-existing diabetes or prediabetes, which is a critical aspect often overlooked in prior studies.

Our study accounts for the occurrence of hyperglycemia over the entire duration of critical illness, rather than just the first 24 hours as seen in previous studies (13), and improves generalizability, as it includes patients regardless of length of stay (12). Availability of BMI and process of care data, which is often absent in similar investigations, allowed for more precise adjustment in our models. Finally, the consistency of results between Cox proportional hazards and Fine and Gray competing‑risk analyses, notably accounting for death, further validates our findings.

Our study is limited by the absence of direct pre‑ and post‑ICU metabolic assessments (e.g., OGTT, C‑peptide, lipid profiles), which precludes causal inferences. Although we adjusted for BMI, it is an indirect proxy that does not reflect body composition or metabolic health nuances such as visceral adiposity or muscle mass. In addition, lacking ethnicity data prevents evaluation of genetic or population‑specific diabetes risk. Finally, although SH is common in sepsis (17), our subgroup analysis by sepsis status lacked power; future research with larger cohorts should enable diagnosis-specific stratification.

Significant exclusions due to missing HbA1c could cause selection bias. We addressed this with two sensitivity analyses: 1) reincorporating patients with missing HbA1c (and using missing HbA1c as a covariate), 2) analyzing a routine HbA1c sampling subgroup. Both preserved the CIAH-diabetes risk association, suggesting minimal selection bias and supporting the robustness of our results.

Observational studies carry the risk of misclassification and unmeasured confounding. Incomplete NDR coverage may underestimate diabetes incidence. We could not adjust for post-ICU factors, such as continued exposure to diabetogenic drugs (e.g., corticosteroids, calcineurin inhibitors [18], or statins [19]), which could affect the incidence of diabetes separately from SH. Furthermore, our use of insulin treatment as proxy readout for SH, whereas practical, has a risk of not accurately reflecting the true glycemic state, as it does not capture when patients were managed outside treatment protocols. The study cohort was derived from a single healthcare system, which could limit generalizability of our findings to other populations or regions with differing ICU practices. However, our blood glucose targets (6–10 mmol/L [108–180 mg/dL]) and insulin administration thresholds are broadly consistent with international guidelines, such as those of the Surviving Sepsis Campaign (20), and the American Diabetes Association (21). In addition, our findings corroborate prior research from Australia, Scotland, and other regions that link SH to post-ICU diabetes risk, suggesting that the underlying biological mechanisms reflected in our results transcend any regional practices. Lastly, the observational design precludes causation, and residual confounding is possible.

CONCLUSIONS

In our cohort of ICU survivors without diabetes and HbA1c below 42 mmol/mol (6%), the requirement of insulin to maintain blood glucose between 6 and 10 mmol/L (108–180 mg/dL) was strongly associated with development of post-ICU diabetes. This finding highlights the potential value of proactive diabetes screening and structured follow-up care for ICU survivors who require insulin during critical illness.

ACKNOWLEDGMENTS

We thank Indir Becic, Business Intelligence Architect, for assisting us with data extraction.

Supplementary Material

ccm-53-e2562-s001.docx ^{(75.1KB, docx)}