Association of 24 hour mean blood glucose levels with mortality in patients with acute pancreatitis

Abstract

This study aims to investigate the impact of 24 hour mean blood glucose (24h MBG) levels on the mortality of patients hospitalized with acute pancreatitis (AP) and to explore the clinical implications of this association. A retrospective examination of clinical datasets of AP patients was performed using the Medical Information Mart for Intensive Care-IV database. The optimal 24h MBG levels threshold was determined by the X-tile tool. Univariate and multivariate logistic regression analyses were used to assess the prognostic significance of 24h MBG levels. A survival analysis was carried out, and Kaplan–Meier survival curves were constructed to show survival probabilities over time. To strengthen result validity and reduce covariate confounding, a 1:1 propensity score matching (PSM) technique was implemented. An external validation was also performed with a cohort of 766 AP patients from the eICU database. A total of 977 subjects were enrolled in this study. The 24h MBG level’s optimal cutoff for predicting in-hospital mortality was determined to be 146 mg/dL. Following PSM analysis, 336 pairs of subjects were identified, achieving equilibrium in most baseline characteristics. Kaplan–Meier survival analysis demonstrated that patients with 24h MBG of 146 mg/dL or higher exhibited significantly elevated rates of in-hospital, 30-day, and 90-day mortality compared to those with 24h MBG below 146 mg/dL (P < .05). Multivariate logistic regression analysis, after adjusting for potential confounding factors, indicated that patients with 24h MBG of 146 mg/dL or higher had a markedly increased odds of in-hospital mortality relative to those with a mean BG below 146 mg/dL (odds ratio = 3.9, 95% confidence interval: 1.86~8.21, P < .001). Consistent results were observed in the PSM-adjusted cohort. Our findings indicate a significant correlation between 24h MBG levels and mortality in patients who have suffered from AP. The results presented in this study are significant and emphasize the necessity for healthcare providers to take into account the impact of 24h MBG levels in effectively managing mortality.

1. Introduction

Acute pancreatitis (AP) is a severe inflammatory condition of the pancreas, which is precipitated by a diverse array of etiological factors that result in the premature activation of pancreatic enzymes.^[1] Studies have shown that the global incidence of AP is 33.74/100,000, and the mortality rate is 1.6/100,000.^[2] Severe AP (SAP) is characterized by its potential to induce necrosis of pancreatic tissue, systemic inflammatory response syndrome, and multiple organ dysfunction syndrome in the early stages of the disease. This condition confers a significant mortality rate, which has been reported to range from 30 to 47%.^[3] The pathophysiological mechanisms underlying SAP involve a complex interplay of inflammatory mediators and cellular responses that lead to the systemic manifestations and high mortality associated with the disease.^[1] Studies have shown that pancreatitis is associated with higher blood glucose levels and a history of diabetes, and that elevated fasting glucose and diabetes are risk factors that damage several organs, including pancreas.^[4,5] Recently, a number of studies have found that blood glucose levels affect the mortality of hospitalized patients.^[6,7] However, the association between 24h mean blood glucose (24h MBG) levels and the mortality in hospitalized patients with AP is not yet determined. Hence, in this study, we aimed to investigate the relationship between 24h MBG levels and mortality in patients with AP in the ICU.

2. Materials and methods

2.1. Data source

All data were extracted from the Medical Information Mart for Intensive Care-IV (MIMIC-IV version 2.2) database which is a large, publicly available electronic health record dataset that contains data from patients admitted to the emergency department or an intensive care unit at the Beth Israel Deaconess Medical Center in Boston, MA.^[8] To gain access to the database, we completed the National Institutes of Health’s web-based course and pass the examinations (certificate record id: 59099686). Since this study involved the analysis of public databases, it was exempt from the need for informed consent from patients and approval from the Institutional Review Board. The eICU program is a multi-center database that enrolled 200,859 patients from 208 hospitals throughout the continental United States in 2014 and 2015.^[9]

2.2. Study population

Using the International Classification of Diseases (ICD), ninth revision (ICD-9, code 577.0) and tenth revision (ICD10, code K85), we identified the patients diagnosed with AP from the MIMIC-IV v2.2 database. We only kept data on the patients’ initial ICU admission for those patients who had multiple admissions. Patients were excluded if they were under 18 years of age, lacked clinical records of blood glucose, or had an ICU stay duration of <24 hours. Finally, 977 patients were enrolled in this research (Fig. 1).

Figure 1.

Flowchart illustrating the selection process of study participants. MIMIC-IV = Medical Information Mart for Intensive Care-IV.

2.3. Data collection

Upon admission, demographic variables, comprising age, gender, and race, were systematically retrieved from the MIMIC-IV database for analysis. Additionally, comorbidities (diabetes, hypertension, chronic pulmonary disease [COPD], congestive heart failure, cirrhosis, hyperlipidemia, and myocardial infarction), scores (SOFA, systemic inflammatory response syndrome, SAPSII, GCS, and CHARLSON), and enteral or parenteral treatment were also included in this analysis. Additionally, the comorbidities mentioned above refer to the chronic underlying diseases that had been diagnosed at the time of admission. More importantly, laboratory indexes and Vital signs on admission were collected. The outcome in this cohort was the mortality of AP patients.

2.4. Statistical analysis

All statistical analyses were performed using R software (version 4.2.2). Continuous variables were first tested for normality using the Shapiro–Wilk test. Normally distributed data are presented as mean ± standard deviation, while non-normally distributed data are presented as median (interquartile range).

For group comparisons, Student t test or the Mann–Whitney U-test was used depending on data distribution. Categorical variables were analyzed using the chi-square test or Fisher exact test, as appropriate.

A generalized additive model (GAM) was employed to assess potential nonlinear relationships between 24h MBG levels and mortality. Logistic regression analyses (both univariable and multivariable) were used to identify independent predictors of in-hospital mortality. Model I adjusted for age, sex, and ethnicity; Model II further adjusted for comorbidities including diabetes, hypertension, COPD, congestive heart failure, cirrhosis, hyperlipidemia, and myocardial infarction; Model III additionally adjusted for laboratory variables such as white blood cell, albumin, total bilirubin, blood urea nitrogen (BUN), and electrolytes (sodium, potassium, and calcium).

Propensity score matching (PSM) was performed at a 1:1 ratio using nearest-neighbor matching without replacement and a caliper width of 0.02 to minimize potential confounding. The covariates used for matching included age, sex, comorbidities, and severity scores (SOFA, SAPSII, and CHARLSON).

Kaplan–Meier (K-M) survival curves were constructed to estimate survival probabilities, and the log-rank test was used to compare survival differences between groups. Statistical significance was set at a two-sided P-value < .05.

3. Results

3.1. Baseline data

We enrolled a total of 977 patients with a mean age of 60.0 ± 17.3 years, among whom 81.7% were white and 58.1% were male. The 24h MBG level in serum of patients during the first 24 hours after admission to the ICU. It may be the average of 2 to 5 measurements taken during this period. The receiver operating characteristic curve of 24h MBG was plotted, and its area under the curve (AUC) was 0.620 (95% confidence interval [CI] 0.552–0.689). The best cutoff value was 146, and the evaluation specificity and sensitivity were 55.0 and 68.7% respectively, and the Youden index was 0.237. Based on the cutoff value, 977 patients were divided into low glucose (24h MBG < 146 mg/dL, n = 641) and high glucose (24h MBG ≥ 146 g/dL, n = 336). The groups categorized by the optimal cutoff value of 24h MBG levels exhibited differences in gender, diabetes, hyperlipidemia, scores, bilirubin, BUN and Scr (all P < .05; Table 1).

Table 1.

Participants’ baseline characteristics.

Characteristic	Total	Low Glucose (<146 mg/dL)	High Glucose (≥146 mg/dL)	P value
Before PSM
N	977	641	336
Demographic
Age, yrs	60.0 ± 17.3	60.1 ± 18.3	59.8 ± 15.2	.76
Gender, n (%)				.005
Female	409 (41.9)	289 (45.1)	120 (35.7)
Male	568 (58.1)	352 (54.9)	216 (64.3)
Race, n (%)				.256
Asian	34 (3.5)	23 (3.6)	11 (3.3)
White	798 (81.7)	535 (83.5)	263 (78.3)
Black	106 (10.9)	60 (9.4)	46 (13.7)
Other	39 (4.0)	23 (3.6)	16 (4.8)
Comorbidities, n (%)
Diabetes	275 (28.1)	117 (18.3)	158 (47.0)	<.001
Hypertension	444 (45.4)	274 (42.7)	170 (50.6)	.904
Chronic pulmonary disease	46 (4.7)	33 (5.1)	13 (3.9)	.37
Congestive heart failure	181 (18.5)	123 (19.2)	58 (17.3)	.462
Cirrhosis	97 (9.9)	63 (9.8)	34 (10.1)	.885
Hyperlipidemia	281 (28.8)	161 (25.1)	120 (35.7)	<.001
Myocardial infarct	171 (17.5)	113 (17.6)	58 (17.3)	.886
Scores
SOFA	5.0 (2.0–8.0)	5.0 (2.0–7.0)	6.0 (3.0–9.0)	<.001
SIRS	3.0 (2.0–3.0)	3.0 (2.0–3.0)	3.0 (3.0–4.0)	<.001
SAPS II	34.0 (24.0–45.0)	32.0 (23.0–42.0)	38.0 (25.0–50.0)	<.001
GCS	15.0 (14.0–15.0)	15.0 (14.0–15.0)	15.0 (14.0–15.0)	.082
CHARLSON	4.0 (2.0–6.0)	4.0 (2.0–6.0)	4.0 (2.0–6.0)	.068
Vital signs
SpO2, %	96.3 (94.9–97.8)	96.2 (94.8–97.7)	96.4 (94.9–97.8)	.599
Temperature, °C	36.9 (36.7–37.3)	36.9 (36.6–37.3)	36.9 (36.7–37.3)	.14
HR, beats/min	93.6 ± 18.1	91.8 ± 17.9	97.1 ± 17.9	<.001
SBP, mm Hg	122.1 ± 19.2	121.5 ± 18.8	123.3 ± 20.1	.166
DBP, mm Hg	69.0 ± 13.9	68.3 ± 13.5	70.4 ± 14.4	.028
RR, times/min	20.1 (17.2–23.3)	19.6 (16.9–22.6)	21.1 (18.0–24.2)	<.001
Laboratory events
WBC, 109/L	12.0 (8.3–17.1)	11.6 (8.3–17.1)	12.1 (8.3–17.1)	.851
Neutrophils, 109/L	11.9 (8.0–16.7)	10.3 (7.0–15.6)	12.2 (8.1–17.3)	.338
Lymphocyte, K/uL	0.9 (0.5–1.5)	0.9 (0.5–1.5)	1.0 (0.5–1.5)	.59
Hb, g/dL	10.8 (9.4–12.4)	10.8 (9.4–12.3)	10.9 (9.5–12.4)	.555
Albumin, g/dL	3.0 (2.5–3.4)	3.0 (2.5–3.5)	3.0 (2.5–3.4)	.799
PT, s	14.4 (12.9–16.9)	14.6 (13.0–17.0)	14.3 (12.9–16.8)	.035
APTT, s	31.2 (27.4–37.3)	31.0 (27.8–37.3)	31.3 (27.2–37.2)	.09
TBil, mg/dL	1.1 (0.6–2.9)	1.2 (0.6–3.4)	1.0 (0.6–2.5)	.451
DBil, mg/dL	2.7 (1.1–5.4)	3.0 (2.1–5.8)	2.4 (0.8–5.4)	.768
IBIL, mg/dL	1.0 (0.6–1.9)	1.0 (0.6–2.1)	1.0 (0.6–1.7)	.855
ALT, IU/L	54.0 (26.0–163.0)	54.0 (25.0–168.5)	54.5 (26.0–151.5)	.01
AST, IU/L	72.0 (36.0–181.8)	74.0 (38.0–182.0)	71.5 (35.0–180.0)	.044
BUN, mg/dL	18.7 (11.7–34.5)	17.1 (10.6–30.4)	20.0 (12.2–37.2)
SCr, mg/dL	1.0 (0.7–1.8)	0.9 (0.7–1.6)	1.1 (0.7–1.9)
In-hospital mortality, n (%)	91 (9.3)	41 (6.4)	50 (14.9)	<.001
Length of ICU stay, days	5.9 ± 9.1	5.1 ± 8.2	7.4 ± 10.6	<.001
After PSM
N	672	336	336
Demographic
Age, yrs	59.9 ± 15.2	60.0 ± 15.2	59.8 ± 15.3	.883
Gender, n (%)				1.000
Female	240 (35.7)	120 (35.7)	120 (35.7)
Male	432 (64.3)	216 (64.3)	216 (64.3)
Race, n (%)				.474
Asian	24 (3.5)	13 (3.9)	11 (3.3)
White	539 (80.2)	276 (82.1)	263 (78.3)
Black	80 (11.9)	34 (10.1)	46 (13.7)
Other	39 (4.0)	23 (3.6)	16 (4.8)
Comorbidities, n (%)
Diabetes	229 (34.1)	71 (21.1)	158 (47.0)	<.001
Hypertension	319 (47.5)	149 (44.3)	170 (50.6)	.105
Chronic pulmonary disease	34 (5.1)	21 (6.2)	13 (3.9)	.159
Congestive heart failure	125 (18.6)	67 (19.9)	58 (17.3)	.372
Cirrhosis	76 (11.3)	42 (12.5)	34 (10.1)	.33
Hyperlipidemia	215 (32.0)	95 (28.3)	120 (35.7)	.039
Myocardial infarct	122 (18.2)	64 (19.0)	58 (17.3)	.548
Scores
SOFA	5.0 (3.0–8.0)	5.0 (2.0–7.0)	6.0 (3.0–9.0)	.005
SIRS	3.0 (2.0–4.0)	3.0 (2.0–3.0)	3.0 (3.0–4.0)	<.001
SAPS II	35.0 (25.0–46.0)	33.0 (25.0–42.2)	38.0 (25.0–50.0)	.001
GCS	15.0 (14.0–15.0)	15.0 (14.0–15.0)	15.0 (14.0–15.0)	.012
CHARLSON	4.0 (2.0–6.0)	4.0 (2.0–6.0)	4.0 (2.0–6.0)	.374
Vital signs
SpO2, %	96.3 (94.9–97.8)	96.2 (94.9–97.8)	96.4 (94.9–97.8)	.721
Temperature, °C	36.9 (36.7–37.3)	36.9 (36.6–37.2)	36.9 (36.7–37.3)	.123
HR, beats/min	94.6 ± 17.9	92.0 ± 17.6	97.1 ± 17.9	<.001
SBP, mm Hg	122.4 ± 19.3	121.5 ± 18.6	123.3 ± 20.1	.356
DBP, mm Hg	69.8 ± 13.8	69.2 ± 13.1	70.4 ± 14.4	.146
RR, times/min	20.3 (17.2–23.6)	19.5 (16.6–22.7)	21.1 (18.0–24.2)	<.001
Laboratory events
WBC, 109/L	11.9 (8.3–17.1)	11.2 (7.9–16.7)	12.8 (9.2–17.7)	.003
Neutrophils, 109/L	11.6 (7.6–16.6)	11.1 (7.1–16.3)	12.1 (8.5–17.0)	.174
Lymphocyte, K/uL	1.0 (0.5–1.5)	0.9 (0.5–1.5)	1.0 (0.5–1.4)	.971
Hb, g/dL	10.8 (9.4–12.4)	10.7 (9.3–12.2)	11.1 (9.6–12.8)	.003
Albumin, g/dL	3.0 (2.5–3.4)	3.0 (2.6–3.5)	3.0 (2.5–3.4)	.26
PT, s	14.4 (12.9–17.0)	14.4 (13.0–17.2)	14.4 (12.9–16.6)	.388
APTT, s	31.3 (27.4–37.4)	31.6 (28.0–38.1)	30.3 (26.9–36.6)	.064
TBil, mg/dL	1.1 (0.6–2.9)	1.2 (0.6–3.3)	1.0 (0.5–2.2)	.004
DBil, mg/dL	2.7 (1.1–5.4)	3.1 (1.6–6.3)	2.0 (0.8–3.0)	.003
IBIL, mg/dL	1.0 (0.6–1.9)	1.0 (0.6–2.4)	0.9 (0.6–1.2)	.038
ALT, IU/L	54.5 (26.0–165.0)	55.0 (25.1–158.5)	54.0 (27.0–172.5)	.765
AST, IU/L	73.2 (36.5–181.8)	70.5 (38.0–179.0)	80.0 (34.5–187.0)	.631
BUN, mg/dL	18.7 (11.7–34.0)	17.0 (10.4–30.0)	23.7 (14.0–39.4)	<.001
SCr, mg/dL	1.0 (0.7–1.8)	0.9 (0.7–1.6)	1.2 (0.8–2.0)	<.001
In-hospital mortality, n (%)	76 (11.3)	26 (7.7)	50 (14.9)	.003
Length of ICU stay, d	6.2 ± 9.1	5.0 ± 7.2	7.4 ± 10.6	.001

ALT = alanine aminotransferase, APTT = activated partial thromboplastin time, AST = aspartate aminotransferase, BUN = blood urea nitrogen, DBil = direct bilirubin, DBP = diastolic blood pressure, GCS = Glasgow coma score, Hb = hemoglobin, HR = heart rate, IBIL = indirect bilirubin, PSM = propensity score matching, PT = prothrombin time, RR = respiratory rate, SAPS II = Simplified Acute Physiology Score II, SBP = systolic blood pressure, SCr = serum creatinine, SIRS = systemic inflammatory response syndrome, SOFA = Sequential Organ Failure Assessment, SpO2 = percutaneous oxygen saturation, TBil = total bilirubin, WBC = white blood cell.

3.2. Prognostic significance of 24-hour blood glucose levels

The GAM analysis demonstrated an apparent U-shaped association between 24h MBG and in-hospital mortality in patients with AP (Fig. 2). The fitted curve suggested that the risk of in-hospital mortality was relatively lower within the intermediate range of 24h MBG and tended to increase toward both lower and higher glucose levels. However, the CIs became wider at the extreme values of 24h MBG, particularly at the lower end of the distribution, indicating greater uncertainty because of the smaller number of observations in these ranges. The K-M curves (Fig. 3) demonstrate that patients with 24h MBG ≥ 146 mg/dL have significantly higher in-hospital mortality, 30-day mortality, and 90-day mortality compared to those with 24h MBG < 146 mg/dL (P < .01).

Figure 2.

Association between 24-hour mean blood glucose (24h MBG) and in-hospital mortality in patients with acute pancreatitis. The solid line represents the fitted association estimated using a generalized additive model, showing an apparent U-shaped pattern. The dashed lines represent the corresponding 95% confidence intervals.

Figure 3.

Kaplan–Meier (K-M) curves indicate the relationship between the 24h MBG and in-hospital mortality (A), 30-day mortality (B), and 90-day mortality (C). Red line: high glucose ≥ 146 mg/dL; blue line: low glucose.

In our study, we employed logistic regression models to ascertain the independent impact of 24h MBG levels on the risk of in-hospital mortality. The analysis was conducted using both univariable and multivariable logistic regression models (Table 2 and Table S1, Supplemental Digital Content 1). In the univariable model, an association was observed between elevated 24h MBG levels and the risk of in-hospital death, with an odds ratio (OR) of 1.01 (95% CI: 1–1.01), which was statistically significant (P < .001). Subsequently, we conducted multivariable analyses with adjustments for various confounders. Model I incorporated age, sex, and ethnicity as covariates. Model II expanded upon Model I by including additional variables such as diabetes, hypertension, COPD, congestive heart failure, cirrhosis, hyperlipidemia, and myocardial infarct. Model III further adjusted for white blood cell, albumin, total bilirubin, BUN, sodium, potassium, and calcium. The results from Model III indicated that patients with 24h MBG levels of 146 mg/dL or higher exhibited a significantly increased risk of in-hospital mortality compared to those with 24h MBG levels below 146 mg/dL. The ORs across the models were as follows: Model I (OR = 2.99, 95% CI: 1.89–4.74, P < .001); Model II (OR = 3.57, 95% CI: 2.17–5.87 P < .001); and Model III (OR = 3.9, 95% CI: 1.86–8.21, P < .001).

Table 2.

Multivariate logistic regression analyses for in-hospital mortality in patients with AP before and after PSM.

Characteristic	Non-adjust model	P-value	Model I	P-value	Model II	P-value	Model III	P-value
	OR (95% CI)		OR (95% CI)		OR (95% CI)		OR (95% CI)
Before PSM
Mean blood glucose (mg/dL)	1.01 (1–1.01)	<.001	1.01 (1–1.01)	<.001	1.01 (1–1.01)	<.001	1.01 (1.01–1.02)	<.001
Low glucose (<146 mg/dL)	1 (Ref)		1 (Ref)		1 (Ref)		1 (Ref)
High glucose (≥146 mg/dL)	2.56 (1.65–3.96)	<.001	2.99 (1.89–4.74)	<.001	3.57 (2.17–5.87)	<.001	3.9 (1.86–8.21)	<.001
After PSM
Mean blood glucose (mg/dL)	1.01 (1–1.01)	.004	1.01 (1–1.01)	.002	1.01 (1–1.01)	<.001	1.01 (1–1.01)	<.001
Low glucose (<146 mg/dL)	1 (Ref)		1 (Ref)		1 (Ref)		1 (Ref)
High glucose (≥146 mg/dL)	2.08 (1.26–3.44)	.004	2.87 (1.56–5.26)	.001	3.68 (1.88–7.2)	<.001	2.08 (1.26–3.44)	<.001

AP = acute pancreatitis, CI = confidence interval, OR = odds ratio, PSM = propensity score matching.

3.3. Cutoff values of 24h MBG and admission blood glucose, and their association with in-hospital mortality

To assess the predictive validity of 24h MBG and admission blood glucose for in-hospital mortality, we performed receiver operating characteristic curve analyses. AUC for mean 24-hour blood glucose and admission blood glucose were 0.620 (95% CI 0.552–0.689) and 0.505 (95% CI: 0.447–0.564), respectively (Fig. 4). The optimal cutoff values were determined to be 146 and 109.5 mg/dL, respectively.

Figure 4.

The ROC curves of the predictive value of 24h MBG and admission BG for in-hospital mortality. 24h MBG = 24 hour mean blood glucose, ROC = receiver operating characteristic.

3.4. Subgroup analysis

Stratified analyses based on age (<65 and ≥ 65 years), sex, and the presence of comorbidities were conducted to evaluate the differences in in-hospital mortality between the 2 cohorts (Fig. 5). An examination of the interaction between 24h MBG levels and all subgroup factors was undertaken, and no significant interaction was observed (P > .05).

Figure 5.

The relationship between 24h MBG and in-hospital mortality in subgroup analysis. 24h MBG = 24 hour mean blood glucose, CI = confidence interval, OR = odds ratio.

3.5. The predictive utility of mean blood glucose post PSM

To address the disparity in baseline characteristics between the 2 groups, our team performed a 1:1 PSM to balance potential confounders, resulting in 336 well-matched pairs (Table 1). After PSM, the high-glucose group had a significantly higher in-hospital mortality rate than the low-glucose group (14.9% vs 7.7%, P = .003) and a longer ICU stay (7.4 ± 10.6 vs 5.0 ± 7.2 days, P < .001; Table 1). The outcomes of multivariable logistic regression analyses in patients posterior to PSM analysis revealed that mean BG ≥ 146 mg/dL was still an independent prediction factor of in-hospital death (Model I: OR = 2.87, 95% CI: 1.56–5.26, P = .001; Model II: OR = 3.68, 95% CI: 1.88–7.20, P < .001; Model III: OR = 2.08, 95% CI: 1.26–3.44, P < .001; Table 2).

3.6. External validation with eICU cohort

We further corroborated our findings through validation studies involving patients with AP sourced from the eICU database. A total of 766 individuals were included in the study (Table S2, Supplemental Digital Content 2). After being PSM matched, 638 AP patients with ICU admission were enrolled (319 in the low-glucose group and 319 in the high-glucose group). The statistically significant difference in survival observed between treatment groups in the K-M survival analysis (P = .04), multivariate analyses revealed that 24h MBG exerted a significant influence on in-hospital mortality among patients with AP (Table S3, Supplemental Digital Content 3 and Fig. S1, Supplemental Digital Content 4).

To ensure the robustness of our findings, multiple analytical methods were performed, including univariate and multivariate logistic regression, GAM modeling, PSM analysis, and external validation using the eICU database. Across the primary and external-validation analyses, elevated 24h MBG was consistently associated with an increased risk of adverse outcomes. These converging results underscore the stability and clinical significance of the observed association.

4. Discussion

In this retrospective cohort study, we investigated the association between 24h MBG and in-hospital mortality among critically ill patients with AP. Our findings showed that 24h MBG was significantly associated with in-hospital mortality when analyzed as both a continuous and a categorical variable. Patients with 24h MBG ≥ 146 mg/dL had a significantly higher risk of in-hospital mortality after adjustment for potential confounders. In addition, 24h MBG demonstrated better prognostic discrimination for in-hospital mortality than admission blood glucose. The consistency of these findings across multivariable regression, PSM, and external validation further supports the robustness of the observed association.

The GAM further demonstrated an apparent U-shaped association between 24h MBG and in-hospital mortality. This finding deserves particular consideration because it suggests that the relationship between glucose levels and mortality may not be purely linear. Both relatively low and elevated 24h MBG levels appeared to be associated with increased mortality risk, whereas an intermediate glucose range was associated with a relatively lower risk. This pattern is broadly consistent with previous observations in critically ill patients with AP. Lu et al reported that mean blood glucose was independently associated with in-hospital mortality and that the lowest mortality occurred when mean glucose was approximately 85 to 100 mg/dL.^[10] Similarly, Wang et al reported that both hypoglycemia and hyperglycemia were associated with prolonged hospital stay in patients with AP.^[11] Previous studies have also shown that admission and peak in-hospital glucose levels are independently and dose-dependently associated with worsening AP severity and mortality.^[7]

Several mechanisms may contribute to the increased risk observed at the higher end of the 24h MBG curve. During the early phase of SAP, systemic inflammatory responses and the release of stress-related hormones may contribute to hyperglycemia and impaired glucose homeostasis.^[12] Hyperglycemia may further aggravate inflammatory responses, oxidative stress, and cellular injury, thereby contributing to organ dysfunction and adverse outcomes.^[13,14] Therefore, elevated 24h MBG may not only reflect disturbances in glucose metabolism but may also serve as a marker of greater physiological stress and disease severity. In contrast, relatively low glucose levels in critically ill patients may represent metabolic instability or treatment- and nutrition-related disturbances rather than a protective state. Taken together, the increased risk at both ends of the curve suggests that disruption of glucose homeostasis, rather than hyperglycemia alone, may be relevant to prognosis in critically ill patients with AP.

Nevertheless, the apparent increase in mortality at the lower end of the curve should be interpreted cautiously. As shown in Figure 2, the CIs widened at extreme 24h MBG values, particularly at the lower end, likely reflecting the smaller number of patients represented in these ranges. Therefore, the present findings do not establish an optimal therapeutic glucose target. Rather, the U-shaped pattern indicates that marked deviations of 24h MBG from an intermediate range may identify patients with greater physiological instability and a higher risk of adverse outcomes. Further prospective studies are needed to determine whether specific glucose ranges have therapeutic implications in patients with AP.

Compared with a single admission glucose measurement, 24h MBG may provide a more representative assessment of early glucose status in critically ill patients. Blood glucose levels at ICU admission can fluctuate considerably because of changes in nutritional intake, catecholamine and cortisol release, insulin administration, and other acute interventions. A single measurement may therefore be particularly sensitive to transient physiological changes. In contrast, averaging multiple glucose measurements during the first 24 hours may reduce the influence of short-term fluctuations and better reflect the overall glycemic state during the early phase of critical illness. Consistent with this hypothesis, 24h MBG showed a higher AUC for predicting in-hospital mortality than admission blood glucose in the present study.

Nutritional support may also influence glucose homeostasis in patients with AP. In our cohort, 95 patients received parenteral nutrition (PN), whereas 184 received enteral nutrition (EN). No significant association between nutritional support and 24h MBG was observed, possibly because 24h MBG was calculated during the first 24 hours after ICU admission, when the influence of subsequent nutritional interventions may have been limited. In critically ill patients with AP, EN is generally preferred over PN when feasible.^[15] Previous evidence suggests that EN is associated with improved clinical outcomes, including fewer infectious complications and shorter hospital stays,^[16] whereas PN remains necessary when EN is contraindicated.^[17] Appropriate glucose monitoring remains important regardless of the route of nutritional support, particularly because hyperglycemia has been associated with oxidative stress, inflammation, and adverse clinical outcomes.^[14]

The present study has several strengths. First, the relatively large sample size allowed us to evaluate the association between 24h MBG and in-hospital mortality after adjustment for multiple potential confounders. Second, the use of complementary statistical approaches, including multivariable logistic regression, generalized additive modeling, and PSM, allowed the association to be examined from different analytical perspectives. Third, the use of an independent multicenter eICU cohort for external validation improved the generalizability of our findings. Moreover, the nonlinear analysis highlighted a potential U-shaped association between 24h MBG and mortality, suggesting that marked deviations from an intermediate glucose range may be clinically relevant.

Several limitations should also be acknowledged. First, because this was a retrospective observational study, residual confounding cannot be completely excluded despite multivariable adjustment and PSM. Second, patients without sufficient blood glucose measurements for calculation of 24h MBG were excluded, which may have introduced selection bias. PSM was used to reduce measured baseline differences between the exposure groups but cannot eliminate selection bias related to missing data or unmeasured confounding. Third, 24h MBG may be influenced by factors such as diabetes, the severity of systemic inflammation, nutritional support, and insulin or other glucose-lowering treatments. Although several relevant covariates were considered in the analyses, detailed information on the timing and intensity of glucose-modifying interventions was not fully evaluated. Finally, the relatively small number of patients at the extreme ends of the glucose distribution resulted in wider CIs in the GAM analysis, particularly at lower 24h MBG levels. Therefore, the apparent U-shaped association requires confirmation in larger prospective cohorts.

5. Conclusion

Our findings demonstrated that the 24h MBG was associated with in-hospital mortality among patients with AP. According to our results, compared with patients with 24h MBG < 146 mg/dL, patients with 24h MBG ≥ 146 mg/dL were at higher risk of in-hospital mortality, 30-day mortality, 90-day mortality, and prolonged ICU stay.

Author contributions

Data curation: Jialin Song.

Methodology: Shengwei Tian.

Resources: Shengwei Tian.