Association between the geriatric nutritional risk index and all-cause mortality in patients with acute pancreatitis in the intensive care unit: a retrospective cohort study

Abstract

Critical illness complicates acute pancreatitis (AP) with acute malnutrition, which increases mortality. The Geriatric Nutritional Risk Index (GNRI) integrates serum albumin and body weight, but its short-term, intermediate-term, and long-term mortality prognostic value in AP admitted to the ICU remains unclear. A retrospective cohort study including 430 adults with a first ICU admission for AP. The primary outcome was 28-day all-cause mortality; secondary outcomes were 90- and 360-day mortality. Multivariable Cox regression, restricted cubic splines (RCS) and Kaplan–Meier analyses were used to assess linear and non-linear associations; effect modification was examined in prespecified subgroups. Median GNRI was 83.4. Each 1-unit increment reduced 28-day mortality by 6% (HR 0.94, 95% CI 0.91–0.97, P = 0.001) with similar effect sizes at 90 and 360 days. RCS revealed a J-shaped curve with a nadir at ≈ 86.8: below this threshold each unit decrease increased risk by 10.4%, whereas above its risk rose by 24.2%. The high-GNRI group had a 62% lower 28-day mortality than the low-GNRI group (HR 0.38, 95% CI 0.21–0.70, P = 0.002). Survival curves remained significantly separated (log-rank P = 0.0018). Subgroup analyses showed stronger protection in males and patients < 60 years (P-interaction < 0.05). GNRI is a rapid, objective and non-linear predictor of short-term, intermediate-term, and long-term mortality in critically ill AP. The J-shaped association supports early nutritional inflammatory risk stratification. Further prospective studies should assess the utility of the GNRI to guide nutritional support interventions in this patient population.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-40767-1.

Introduction

Acute pancreatitis (AP) is a common abdominal condition characterized by pancreatic self-digestion and systemic inflammation resulting from abnormal enzyme activation¹. It typically manifests as severe abdominal pain and organ dysfunction². Worldwide statistics show an increasing yearly occurrence of AP, ranging from 13 to 45 cases per 100,000 individuals³. Approximately 20% of AP cases progress to severe acute pancreatitis (SAP), which has an in-hospital mortality rate of 30% to 40% due to multiple organ failure^4–6. The clinical course of AP exists on a severity continuum, with clinical nutrition playing a dual role across the entire spectrum. In milder cases, pre-existing malnutrition may compromise recovery and increase complication risks; in severe cases, the hypermetabolic state induced by critical illness leads to rapid protein catabolism, negative energy balance, and accelerated muscle wasting, resulting in malnutrition as a disease complication. Approximately 68.2% of SAP patients exhibit nutritional biomarker irregularities upon admission. Malnutrition is associated with heightened risks of secondary infections, prolonged hospital stays, and elevated mortality rates^7,8.This bidirectional relationship, where malnutrition is both a potential risk factor present at onset and a consequence of disease progression, highlights the critical need for early nutritional risk assessment in all AP patients, not only those with established severe disease.

The American Society for Parenteral and Enteral Nutrition (ASPEN) emphasizes the importance of early nutritional risk assessments to improve outcomes for critically ill patients⁹. The Nutritional Risk Screening 2002 (NRS 2002) is an evidence-based tool widely used in clinical environments to assess nutritional risks in hospitalized patients. However, its application in ICU patients is notably restricted due to challenges posed by critical illness, such as mechanical ventilation or altered mental status, which hinder effective communication during assessments^10,11. Geriatric Nutritional Risk Index (GNRI) objectively measures serum albumin, weight and height; its combination allows it to concurrently indicate a person’s protein reserves and energy metabolism status. Studies have proven that the GNRI is effective in predicting death and unfavorable results in a range of clinical contexts, with the ICU being one of them^12–14. However, the prognostic value and optimal cutoff of GNRI specifically in AP patients—encompassing the full severity spectrum—remain undefined.

Recent studies emphasize the importance of early nutritional intervention for AP, particularly in severe cases¹⁵. Utilizing validated tools or timely nutritional assessment helps identify high-risk patients, enabling the creation of tailored nutrition support strategies¹⁶. This study investigates the relationship between the GNRI and the 28-day mortality rate in ICU-admitted AP, utilizing the MIMIC-IV database. The study proposes that a reduced GNRI correlates with heightened mortality risk and negative outcomes in these patients.

Materials and methods

Data sources

The study exclusively utilized patient data from the MIMIC-IV (v3.1) database, a comprehensive public resource overseen by MIT’s Computational Physiology Laboratory. The database comprises hospitalization records from Boston Beth Israel Deaconess Medical Center (BIDMC) spanning 2008 to 2022. All personal data is anonymized, and a random code is used instead of a patient identifier to ensure privacy. As a result, BIDMC does not need ethics approval or informed consent. The MIMIC-IV (v3.1) database is available on Physio Net. The principal investigator obtained CITI Program certification (Record ID: 65963592) through successful completion of required modules in research ethics and data stewardship, including Conflicts of Interest in Human Subjects Research and Secondary Analysis of Existing Datasets, thereby securing institutional authorization for the extraction of de-identified clinical records from the designated health information repository.

Study population

Hospitalization data for AP were extracted using ICD-9 code 577.0 and ICD-10 codes K8590, K8510, K8520, K859, K8580, K8591, K851, K8521, K8500, K852, K8592, K8511, K8530, K8581, K858, K8522, J0140, K853, K8512, K8582, K8501, K8502, K850, and K8531.GNRI is a rapid and objective assessment method. This study focuses on AP patients aged over 18 who are admitted to the ICU for the first time. The data used in this study are derived from the patients’ hospitalization period in the ICU for the first time. Patients were excluded if they lacked albumin, height, or weight data, stayed in the ICU for less than 24 h, had malignant tumors, or had incomplete or inaccurate follow-up data. The final study cohort consisted of 430 patients. These patients were subsequently divided into two groups based on the median value: the High GNRI group (≥ 83.4) and the Low GNRI group (<83.4) (Fig. 1).

Fig. 1.

Flowchart of patient selection.

Data collection

SQL was utilized for covariate extraction, employing Navicat Premium (v16) and PostgreSQL (v13.7.2) as tools. The study identified five main covariate components: laboratory variables, comorbidities, organ dysfunction scores, vital signs, and demographic characteristics, detailed in Table 1. clinical nutrition was assessed using the GNRI, originally developed by Bouillanne as a malnutrition screening tool and mortality predictor for hospitalized elderly patients¹⁷. The GNRI was calculated using the following formula: GNRI = 14.89 × serum albumin in grams per deciliter + (41.7 × actual body weight/ideal body weight), The ideal body weight (WLo) was estimated using the Lorentz formula, which takes into account height (H) and sex, For men: , For women: , When the actual body weight surpassed WLo, the weight-to-WLo ratio was standardized to 1.

Table 1.

Characteristics of the study population according to GNRI groups.

Variables	Total (n = 430)	GNRI		P value
		Low GNRI (n = 202)	High GNRI (n = 228)
Sex, n (%)
Male	252 (58.6)	117 (57.9)	135 (59.2)	0.786
Female	178 (41.4)	85 (42.1)	93 (40.8)
Age	56.7 ± 17.6	57.4 ± 17.0	56.0 ± 18.2	0.43
Weight	90.2 ± 23.6	89.1 ± 23.0	91.2 ± 24.1	0.362
Height	169.6 ± 10.1	169.5 ± 10.5	169.7 ± 9.9	0.882
HR	87.8 ± 15.7	91.2 ± 16.6	88.6 ± 15.5	< 0.001
RR	21.7 ± 4.4	22.3 ± 4.6	21.7 ± 4.5	0.008
Spo2, %	96.1 ± 4.3	96.1 ± 4.6	96.1 ± 4.0	0.917
T, °C	37.0 ± 0.9	37.0 ± 1.1	37.0 ± 0.8	0.624
Laboratory parameters
WBC, 109/L	15.2 ± 8.2	15.4 ± 8.3	14.9 ± 8.2	0.514
RBC, 109/L	3.7 ± 0.9	3.5 ± 0.8	3.8 ± 1.0	< 0.001
PLT, 109/L	213.2 ± 129.1	222.2 ± 146.9	205.2 ± 110.8	0.173
HB, g/dL	11.2 ± 2.7	10.7 ± 2.4	11.7 ± 2.8	< 0.001
RDW, %	15.1 ± 2.3	15.4 ± 2.3	14.9 ± 2.3	0.021
ALB, g/dL	2.8 ± 0.6	2.3 ± 0.4	3.3 ± 0.4	< 0.001
Na+, mEq/L	138.1 ± 6.4	137.3 ± 7.1	138.8 ± 5.7	0.023
K+, mEq/L	4.2 ± 0.8	4.1 ± 0.8	4.3 ± 0.9	0.123
Ca2+, mg/dL	7.7 ± 1.2	7.4 ± 1.2	8.0 ± 1.1	< 0.001
AG, mEq/L	16.1 ± 5.5	15.5 ± 5.7	16.7 ± 5.2	0.022
PT, sec	17.2 ± 8.5	17.8 ± 9.2	16.7 ± 7.9	0.166
INR	1.6 ± 0.9	1.6 ± 0.9	1.5 ± 0.8	0.156
TBil, mg/dL	1.1 (0.6, 3.0)	1.0 (0.5, 2.8)	1.2 (0.6, 3.3)	0.164
ALT, IU/L	47.0(22.0,135.8)	37.5 (20.0, 98.5)	61.5(26.8,185.2)	0.001
AST, IU/L	76.0(36.0, 80.5)	70.5(34.0, 71.2)	78.0(40.0, 189.5)	0.203
BUN, mg/dL	22.0 (12.0, 38.0)	24.0 (13.2, 40.8)	20.0(12.0, 35.2)	0.099
Blood Lipase, mg/dL	242.5(74.2, 57.2)	193.0(69.2, 4.0)	288.5(83.5,901.5)	0.018
Cr, mg/dL	1.2 (0.7, 2.1)	1.4 (0.8, 2.3)	1.1 (0.7, 1.8)	0.238
Comorbidities, n (%)
Hypertension	187 (43.5)	82 (40.6)	105 (46.1)	0.254
T2DM	104 (24.2)	46 (22.8)	58 (25.4)	0.519
HF	67 (15.6)	27 (13.4)	40 (17.5)	0.233
CKD	61 (14.2)	32 (15.8)	29 (12.7)	0.354
Hepatitis	69 (16.0)	36 (17.8)	33 (14.5)	0.345
HLP	108 (25.1)	48 (23.8)	60 (26.3)	0.542
Scoring
SOFA	7.3 ± 4.4	7.8 ± 4.3	6.8 ± 4.3	0.01
APSIII	61.2 ± 25.8	68.0 ± 26.5	55.2 ± 23.6	< 0.001
Oasis	36.7 ± 9.4	38.9 ± 8.7	34.9 ± 9.5	< 0.001
Charlson	3.0 (1.0, 5.0)	3.0 (2.0, 5.0)	3.0 (1.0, 6.0)	0.683
Outcomes
ICU stay	23.2 ± 10.3	25.0 ± 12.3	24.0 ± 10.7	< 0.001
Hospital stays	32.0 ± 19.0	35.4 ± 23.6	34.2 ± 18.4	< 0.001
ICU mortality	58 (13.5)	38 (18.8)	20 (8.8)	0.002
28-day mortality	67 (15.6)	43 (21.3)	24 (10.5)	0.002
90-day mortality	98 (22.8)	57 (28.2)	41 (18)	0.012
360-day mortality	115 (26.7)	63 (31.2)	52 (22.8)	0.05

Abbreviations: AG, anion gap; ALB, albumin; ALT, alanine aminotransferase; APSIII, Acute Physiology Score III; AST, aspartate aminotransferase; Blood Lipase, serum lipase; BUN, blood urea nitrogen; Ca²⁺, calcium; Charlson, Charlson Comorbidity Index; CKD, chronic kidney disease; Cr, creatinine; HB, hemoglobin; HF, heart failure; Hepatitis, hepatitis diagnosis; HLP, hyperlipidemia; HR, heart rate; Hypertension, hypertension diagnosis; ICU 28 dead, death within 28 days of ICU admission; ICU 90 dead, death within 90 days of ICU admission; ICU 360 dead, death within 360 days of ICU admission; ICU dead, death during ICU stay; INR, international normalized ratio; K⁺, potassium; LOS ICU day, length of stay in ICU; LOS hospital day, length of stay in hospital; Na⁺, sodium; Oasis, Oxford Acute Severity of Illness Score; PLT, platelet; PT, prothrombin time; RBC, red blood cell; RDW, red-cell distribution width; RR, respiratory rate; Sepsis, sepsis diagnosis; SOFA, Sequential Organ Failure Assessment; SpO₂, pulse oxygen saturation; T, body temperature; T2DM, type 2 diabetes mellitus; TBil, total bilirubin; WBC, white blood cell.

Outcome variables

The primary endpoint of this study was 28-day mortality during ICU stay; the secondary endpoints were mortality at 90 and 360 days. Follow-up was censored at 360 days after ICU admission.

Statistical analysis

The Shapiro-Wilk test was employed to evaluate the normality of data distribution. Continuous data with a normal distribution are shown as mean and standard deviation (SD), whereas skewed data are represented by median and interquartile range (IQR). Numbers and percentages indicate categorical variables. Continuous variables were analyzed using ANOVA or rank sum tests, while categorical variables were compared with χ² or Fisher’s exact tests. Additionally, three different configurations of multivariate Cox regression models were constructed. Variable selection was based on the variables with P-values less than 0.05 in the univariate analysis (Table S1), and the collinearity issues of these variables were considered. Variables with variance inflation factors greater than 10 (such as PT, VIF = 71.028; INR, VIF = 69.458) were excluded (Table S2). The independent relationship between GNRI and 28,90, and 360-day mortality was examined through multivariate Cox regression analysis across three hierarchical models. These models were designed to incrementally adjust for confounding variables, with Model 1 serving as the baseline non-adjusted analysis. Model 2 incorporated demographic adjustments for age and sex, whereas Model 3 provided comprehensive adjustments for age, sex, chronic kidney disease (CKD), hepatitis, sequential organ failure assessment (SOFA), acute physiology score III (APSIII), oxford acute severity of illness score (OASIS), Charlson comorbidity index, red blood cell (RBC), platelet (PLT), hemoglobin (HB), anion gap (AG), sodium (Na⁺), total bilirubin, and blood urea nitrogen (BUN). The study utilized Restricted cubic splines analysis (RCS) models to investigate threshold effects and nonlinear associations between GNRI levels and mortality risk. Survival curves were generated using cumulative risk, Kaplan-Meier, and log-rank analyses to compare mortality across groups with varying GNRI levels at ICU admission. The reproducibility of findings was evaluated through subgroup analyses stratified by comorbidities, age (under 60 vs. 60 and above), and sex. Multiple imputations addressed missing values in continuous variables. All the analyses were performed with the statistical software packages R (http://www.R-project.org, The R Foundation) and Free Statistics software versions 2.3.

Results

Baseline characteristics

Baseline characteristics from 430 patients were analyzed according to GNRI (see Table 1) and stratified them into a high-GNRI group (n = 228) and a low-GNRI group (n = 202) to systematically evaluate demographics, laboratory parameters, comorbidities, and outcomes. Except for sex, age, body weight, height, SpO₂, body temperature, WBC, PLT, electrolytes (K⁺, Na⁺), coagulation indices, aspartate aminotransferase (AST), bilirubin, creatinine, and common comorbidities (hypertension, T2DM, HF, CKD, hepatitis, hyperlipidaemia), no significant differences were observed. Patients with low GNRI (≤ 83.4, n = 202) exhibited a pronounced “high-stress, high-inflammation, low-reserve” profile: heart rate (91.2 ± 16.6 vs. 88.6 ± 15.5 bpm, P < 0.001), respiratory rate (22.3 ± 4.6 vs. 21.7 ± 4.5 breaths/min, P = 0.008) and RDW (15.4 ± 2.3 vs. 14.9 ± 2.3%, P = 0.021) were all higher, whereas RBC count (3.5 ± 0.8 vs. 3.8 ± 1.0 × 10¹²/L, P < 0.001), HB (10.7 ± 2.4 vs. 11.7 ± 2.8 g/dL, P < 0.001), serum calcium (7.4 ± 1.2 vs. 8.0 ± 1.1 mg/dL, P < 0.001) and albumin (2.3 ± 0.4 vs. 3.3 ± 0.4 g/dL, P < 0.001) were significantly lower than in the high-GNRI group. Severity scores (SOFA, APSIII, OASIS) followed the same trend (all P ≤ 0.01). Clinically, low-GNRI patients had longer ICU stays (25.0 ± 12.3 vs. 24.0 ± 10.7 days, P < 0.001), doubled ICU mortality (18.8% vs. 8.8%, P = 0.002) and persistently higher cumulative mortality at 28, 90 and 360 days (21.3% vs. 10.5%; 28.2% vs. 18.0%; 31.2% vs. 22.8%; all P < 0.05).

Multivariable Cox regression analysis

Table 2 demonstrates that GNRI was independently and consistently associated with lower mortality at 28, 90, and 360 days, whether analyzed as a continuous or categorical variable. In the continuous model, each 1-unit increase in GNRI reduced the 28-day risk by 6% (Model-3 h 0.94, 95%CI 0.91–0.97, P = 0.001), with identical effect sizes at 90 and 360 days. Dichotomized analysis revealed a 62% lower 28-day mortality in the high- versus low-GNRI group (HR 0.38, 95%CI 0.21–0.70, P = 0.002). Tertile analysis showed a significant dose–response trend (P = 0.011), with the highest tertile (Q3) experiencing a 60% risk reduction compared with the lowest (Q1) at 28 days (HR 0.4, 95%CI 0.2–0.83, P = 0.013), and this protective effect remained significant at 90 and 360 days.

Table 2.

Multivariate COX regression analysis to assess the association between GNRI and 28-day, 90-day, and 360-day all-cause mortality in acute pancreatitis patients.

Variable	Model 1		Model 2		Model 3
	HR (95%CI)	P	HR (95%CI)	P	HR (95%CI)	P
28-day all-cause mortality
GNRI	0.95(0.93 ~ 0.97)	< 0.001	0.95 (0.92 ~ 0.97)	< 0.001	0.94 (0.91 ~ 0.97)	0.001
Quantiles
Low GNRI	Ref		Ref		Ref
High GNRI	0.46(0.28 ~ 0.76)	0.002	0.45 (0.28 ~ 0.75)	0.002	0.38 (0.21 ~ 0.7)	0.002
Quantiles
Q1	Ref		Ref		Ref
Q2	0.39 (0.22 ~ 0.7)	0.002	0.39 (0.22 ~ 0.71)	0.002	0.38 (0.15 ~ 0.57)	< 0.001
Q3	0.4 (0.22 ~ 0.71)	0.002	0.4 (0.22 ~ 0.72)	0.002	0.4 (0.2 ~ 0.83)	0.013
90-day all-cause mortality
GNRI	0.96(0.94 ~ 0.98)	< 0.001	0.96 (0.94 ~ 0.98)	< 0.001	0.95 (0.93 ~ 0.98)	0.001
Quantiles
Low GNRI	Ref		Ref		Ref
High GNR	0.58(0.39 ~ 0.87)	0.008	0.96 (0.94 ~ 0.98)	< 0.001	0.5 (0.31 ~ 0.8)	0.003
Quantiles
Q1	Ref		Ref		Ref
Q2	0.47(0.29 ~ 0.76)	0.002	0.46 (0.29 ~ 0.75)	0.002	0.37 (0.21 ~ 0.63)	< 0.001
Q3	0.52(0.32 ~ 0.83)	0.006	0.51 (0.32 ~ 0.82)	0.005	0.52 (0.3 ~ 0.91)	0.022
360-day all-cause mortality
GNRI	0.96(0.95 ~ 0.98)	0.001	0.96 (0.94 ~ 0.98)	< 0.001	0.96 (0.93 ~ 0.98)	0.001
Quantiles
Low GNRI	Ref		Ref		Ref
High GNR	0.66(0.46 ~ 0.96)	0.029	0.65 (0.45 ~ 0.94)	0.021	0.59 (0.39 ~ 0.9)	0.015
Quantiles
Q1	Ref		Ref		Ref
Q2	0.56(0.36 ~ 0.88)	0.011	0.56 (0.36 ~ 0.87)	0.01	0.47 (0.26 ~ 0.76)	0.002
Q3	0.58 (0.37 ~ 0.9)	0.014	0.57 (0.36 ~ 0.88)	0.012	0.57 (0.34 ~ 0.95)	0.032

HR: Hazard Ratio, CI: Confidence Interval.

Model 1: Crude.

Model 2: Adjust for Age and Sex.

Model 3: Adjust for Age, Sex, CKD, Hepatitis, SOFA, APSIII, Oasis, Charlson, RBC, PLT, HB, AG, Na+, Bilirubin total, BUN.

Restricted cubic splines analysis

The association between GNRI and mortality risk in patients with AP was examined through RCS analysis. Figure 2A (28-day mortality rate), Fig. 2B (90-day mortality rate), and Fig. 2C (360-day mortality rate) all have undergone adjustments for confounding factors such as age, gender, chronic kidney disease, hepatitis, SOFA score, APSIII score, OASIS score, Charlson comorbidity index, red blood cells, platelets, hemoglobin, prothrombin time, international normalized ratio, uric acid, total bilirubin, and urea nitrogen. Restricted cubic splines (Fig. 2) revealed a uniform J-shaped relationship across all time points: Panel A (28-day mortality) showed an overall P < 0.001 and non-linear P = 0.002, with a nadir at GNRI ≈ 86.8 (Table S3); each unit increase below this threshold reduced the hazard by 10.4%, whereas risk rebounded by 24.2% above it. Panels B (90-day) and C (360-day) mortality recapitulated the same pattern (overall P < 0.001, non-linear P = 0.007 each), indicating a stable, non-linear dose–response between GNRI and short-term, intermediate-term, and long-term all-cause mortality.

Fig. 2.

Restricted cubic spline showing J-shaped relationship between GNRI and 28-day (A), 90-day (B), and 360-day (C) all-cause mortality.

Kaplan–Meier analysis

Kaplan–Meier survival curves (Fig. 3) demonstrated sustained separation between GNRI groups: in Panel A (28-day mortality) the high-GNRI cohort maintained a survival probability of approximately 90% versus 75% in the low-GNRI cohort (log-rank P = 0.0018). This divergence persisted in Panel B (90-day) and Panel C (360-day), with log-rank P values of 0.0075 and 0.028, respectively, indicating that a higher GNRI at ICU admission is significantly associated with reduced short-, intermediate-, and long-term all-cause mortality in patients with acute pancreatitis.

Fig. 3.

Kaplan–Meier curves for 28-day (A), 90-day (B), and 360-day (C) all-cause mortality by GNRI group (median split).

Subgroup analyses

Subgroup analyses presented in Fig. 4 demonstrated that the protective effect of GNRI (per 1-unit increase) on 28-day mortality was consistent across most strata; however, significant effect modification was observed. Males exhibited an HR of 0.88 (95% CI 0.84–0.93), whereas females had an HR of 0.99 (95% CI 0.94–1.04), with a P-for-interaction < 0.001. Patients aged < 60 years derived greater benefit (HR 0.87; 95% CI 0.81–0.94) than those aged ≥ 60 years (HR 0.96; 95% CI 0.92–1.01; P-interaction = 0.023). No significant heterogeneity was detected for hypertension, type 2 diabetes, hepatitis, or hyperlipidemia (all P > 0.05). In Figure S1, Panels A (90-day) and B (360-day) recapitulated this pattern: the survival advantage remained pronounced in males and in patients without chronic kidney disease (CKD), while it was attenuated in females. Notably, the CKD subgroup showed a trend toward greater benefit at 90 and 360 days (P-interaction = 0.041 and 0.026, respectively). Collectively, GNRI maintains stable prognostic value throughout the entire follow-up period, but the magnitude of protection is modified by sex, age, and renal function. As shown in (Fig. 4 and Figure S1).

Fig. 4.

Forest plot of GNRI on 28-day mortality in patients with acute pancreatitis across subgroups.

Discussion

This study enrolled 430 critically ill ICU patients with AP and systematically investigated the association between GNRI and all-cause mortality at 28, 90, and 360 days. This demonstrated that GNRI serves as an independent predictor of short-, intermediate-, and long-term mortality in critically ill patients with severe AP. Using 28-day mortality as the primary endpoint, we found that each one-unit increase in GNRI was associated with a sustained 6% reduction in death risk, and this association remained robust even after full adjustment for age, sex, comorbidities, and organ failure severity (HR 0.94, 95% CI 0.91–0.97, P = 0.001). When GNRI was analysed as a binary variable, the low-GNRI group exhibited a 62% higher 28-day mortality risk; tertile analysis further confirmed that the highest GNRI tertile (Q3) had a 60% lower death risk compared with the lowest tertile, with a significant dose-response trend (P = 0.011). Kaplan-Meier curves visually demonstrated that the high-GNRI group maintained a significant survival advantage throughout the entire 360-day follow-up. A novel finding of the study was the significant J-shaped non-linear relationship between GNRI and mortality, revealed by RCS analysis and consistently observed at 28, 90, and 360 days (overall and non-linear P < 0.001). The inflection point was approximately 86.8; below this value, each unit increase reduced mortality risk by 10.4%, Above the threshold, the association between GNRI and mortality lost statistical significance, suggesting a ceiling effect where GNRI no longer discriminates mortality risk among well-nourished patients. This complex relationship underscores that GNRI serves primarily as a one-directional risk stratification tool—effective at identifying high-risk, malnourished patients who require immediate intervention, but ineffective at further prognostic discrimination among patients with normal nutritional status. This has important clinical implications for resource allocation and intervention prioritization.

Forest plots showed that the protective effect of GNRI on 28-day mortality was consistent across most prespecified subgroups, but the magnitude was significantly modified by sex and age: males experienced a pronounced protective effect (HR 0.88), while females showed virtually no benefit (HR 0.99; P-interaction < 0.001); patients < 60 years gained greater benefit (HR 0.87), which attenuated in those ≥ 60 years (HR 0.96).These sex and age differences persisted at 90 and 360 days. Although GNRI was initially validated for geriatric populations, the findings support its broader applicability in general adult critical care settings. This extension is mechanistically plausible given the index’s reliance on objective physiological parameters (serum albumin, weight, height)—rather than age-specific variables—which reflect universal malnutrition pathophysiology. These results align with recent validations in general adult ICU populations^18–20. Moreover, non-CKD patients derived larger benefits at later time-points, whereas CKD patients also exhibited significant protective effects at 90 and 360 days. In conclusion, the study establishes GNRI as a powerful and independent predictor of mortality across multiple time points in critically ill patients with severe AP. This result is in agreement with earlier research, like the work done by Li et al.¹⁹. identified GNRI as an independent predictor of 28-day mortality in elderly sepsis patients and found it significantly associated with in-hospital mortality in acute kidney injury patients¹⁴. While previous studies have mainly focused on elderly or chronically ill patients, this research uniquely applies GNRI to AP, a condition characterized by severe inflammation, thus addressing a gap in the existing literature.

The prognostic significance of GNRI likely stems from its comprehensive evaluation of clinical nutrition and inflammation. AP often faces pancreatic necrosis, SIRS, and metabolic imbalances, leading to a hypercatabolic state that exacerbates protein loss and hypoalbuminemia²¹. Serum albumin, integral to GNRI, functions both as a traditional nutritional marker and a negative acute-phase protein. In AP, inflammatory substances such as interleukin-6 (IL-6) and tumor necrosis factor-alpha (TNF-α) reduce the liver’s production of albumin and encourage its seepage into surrounding tissues because of increased capillary permeability²². Consequently, hypoalbuminemia indicates both malnutrition and the extent of systemic inflammation. Furthermore, a decreased weight may indicate reduced muscle mass and inadequate energy reserves, compromising the body’s capacity to withstand acute stress²³. The integration of albumin in GNRI with height and weight reflects the bidirectional interaction between metabolic dysfunction and immune activation. In AP, pancreatic necrosis releases damage-associated molecular patterns, which sustain systemic inflammatory response syndrome and elevate oxidative stress. This inflammatory milieu accelerates muscle proteolysis via ubiquitin-proteasome and autophagy-lysosome pathways, further depleting lean body mass and circulating albumin²⁴. Concurrently, hypoalbuminemia exacerbates endothelial dysfunction and fluid shifts, worsening organ perfusion and multi-organ failure risk. By integrating albumin and body weight metrics²⁵, GNRI offers a comprehensive assessment of the nutritional-inflammatory imbalance in AP patients, thereby effectively predicting their prognosis.

While GNRI is widely recognized for its prognostic value in critically ill patients, its effectiveness in AP patients may be influenced by their distinct characteristics. AP pathophysiology involves widespread inflammation, cytokine release, microcirculatory issues, and multi-organ failure²⁶, where a low GNRI could indicate both nutritional deficits and a severe systemic inflammatory response. Thus, GNRI might reflect the clinical nutrition and inflammation severity. Additionally, AP patients experience unique metabolic changes, including insulin resistance, hyperglycemia, increased lipolysis, and potential bacterial translocation due to intestinal barrier dysfunction²⁷. Early enteral nutrition (EN) is recommended to improve intestinal function and decrease infectious complications, although its efficacy may vary based on the patient’s clinical nutrition²⁸. The study indicates that patients with a low GNRI experience significantly prolonged hospital and ICU stays (P < 0.001), possibly due to insufficient nutritional support. While GNRI might help identify those who could benefit from enteral nutrition, additional studies are required to substantiate its effectiveness.

While MIMIC-IV provides valuable real-world critical care data, its retrospective design inherently limits nutritional record standardisation, constraining the analysis to the database’s original variables and documentation practices. Although GNRI demonstrated robust prognostic performance under these constraints, the study did not directly compare it with established nutritional assessment tools such as mNUTRIC or NUTRIC scores²⁹. Future research should prioritise this comparative validation to establish context-specific superiority.

This study confirms a robust GNRI–mortality association in AP ICU patients. However, retrospective data from a single centre (MIMIC-IV) may introduce selection/information biases and limit generalizability. Despite rigorous adjustment for demographic covariates, the single-centre design restricts extrapolation to populations with divergent ethnic distributions or healthcare resource allocations; multicentre validation thus remains essential for clinical implementation. Furthermore, although the primary analysis achieved sufficient power, subgroup estimates with confidence intervals crossing unity require validation in larger cohorts.

The single-timepoint GNRI assessment captures baseline risk stratification but not longitudinal nutritional trajectories, warranting serial measurements in subsequent research to evaluate temporal outcome relationships. Although the admission year was adjusted for in multivariable models, the extended study period (2008–2022) may incompletely capture evolving practice patterns, underscoring the need for contemporary validation amid critical care nutrition advancements.

The prognostic focus minimized therapeutic confounding, yet the omission of personalized nutritional interventions (e.g., enteral/parenteral nutrition) represents a significant constraint. Future multicentre prospective studies should validate GNRI ‘s predictive value and assess nutritional intervention impacts while addressing these temporal and methodological dimensions.

The study identified a nonlinear J-shaped relationship between GNRI and 28-day mortality in ICU patients with AP, emphasizing the critical role of evaluating clinical nutrition in the management of AP and establishing a standard for personalized nutritional interventions. Although limited by its retrospective nature, the study’s results hold substantial clinical relevance and warrant further validation and refinement through prospective research. In the future, GNRI could be integrated into a comprehensive assessment system for AP patients, facilitating early identification of high-risk groups and improving prognosis.

Conclusion

In this large ICU cohort, GNRI demonstrates a stable, J-shaped association with short- and long-term mortality in critically ill patients with acute pancreatitis, with an optimal inflection point around 86.8. Its protective effect is most pronounced in males and those aged < 60 years. While incorporating GNRI into early ICU assessment may facilitate timely risk stratification and inform individualized nutritional strategies, whether this approach will improve outcomes needs to be tested in future interventional studies.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

C.W, K.Z. and Y.L.: Conceptualization, Investigation, Methodology, Resources, Data curation, Formal Analysis, Software, Project administration, Writing – original draft, Writing – Review & Editing.C.W, X.L., X.T. and Q.Z.: Methodology, Data curation, Resources, Software, Validation, Visualization, Investigation, Supervision, Writing – original draft.All authors have seen this version of the manuscript and approved it for submission.

Data availability

This study employed publicly accessible datasets available at [https://mimic.mit.edu/](https:/mimic.mit.edu) .

Declarations

Competing interests

The authors declare no competing interests.