Abstract
Abstract
Background
The prognostic value of the geriatric nutrition risk index (GNRI) in critically ill patients with heart failure (HF) remains unclear, despite its established association with mortality in critically ill patients with HF.
Objective
To investigate the association between GNRI and mortality and evaluate the additive value of GNRI for predicting mortality in critically ill patients with HF.
Methods
This retrospective cohort study analysed 4058 patients aged ≥65 years with HF from the Medical Information Mart for Intensive Care IV database. GNRI was categorised into GNRI >98 (no malnutrition risk) and GNRI ≤98 (at risk of malnutrition). Associations between GNRI and in-hospital, 30-day and 1-year mortality were evaluated using logistic regression, Cox proportional hazards models and restricted cubic spline analyses. Subgroup and sensitivity analyses assessed robustness, while receiver operating characteristic curves compared the predictive performance of Sequential Organ Failure Assessment (SOFA) scores with and without GNRI.
Results
The overall mortality rates were 27.1% (in-hospital), 32.8% (30-day) and 53.4% (1-year). Patients with GNRI >98 had significantly better survival odds for in-hospital mortality (OR=0.70, 95% CI 0.61 to 0.80) and higher mortality risks for 30-day (HR=0.67, 95% CI 0.60 to 0.75) and 1-year (HR=0.67, 95% CI 0.61 to 0.73) outcomes, even after adjusting for confounders. Restricted cubic spline analysis revealed an L-shaped association between GNRI and in-hospital/1-year mortality, while a linear relationship was found for 30-day mortality. Adding GNRI to SOFA scores was significantly associated with enhanced predictive accuracy across all outcomes (areas under the curve (AUCs)=0.724, 0.724, 0.706 and 0.670 for in-hospital, 30-day and 1-year mortality, respectively).
Conclusions
GNRI is associated with mortality in critically ill patients with HF and improves SOFA score accuracy, supporting its integration into routine assessments to enhance risk stratification and inform clinical decision-making.
Keywords: Heart failure, Mortality, Frail Elderly
STRENGTHS AND LIMITATIONS OF THIS STUDY.
This study used a large, well-characterised intensive care unit (ICU) database (the Medical Information Mart for Intensive Care IV) with standardised data collection protocols and predefined inclusion criteria.
Multiple statistical approaches were employed, including Cox proportional hazards regression, Kaplan–Meier survival analysis and restricted cubic splines to assess mortality associations.
Model performance was evaluated using discrimination metrics (AUC), calibration assessment and sensitivity analyses with multiple imputation for missing data.
The retrospective single-centre design limits external generalisability and may introduce residual confounding despite comprehensive statistical adjustments.
Nutritional assessment using geriatric nutrition risk index was conducted only at ICU admission without longitudinal monitoring, potentially missing dynamic nutritional changes during hospitalisation.
Introduction
Heart failure (HF) is a major global health challenge, affecting over 64 million individuals worldwide, with a predominant burden on older populations and resource-limited regions.1 2 Malnutrition has become a critical determinant of outcomes due to metabolic disruptions,3 reduced appetite4 and systemic inflammation5 caused by HF, particularly in intensive care unit (ICU) settings.6 7 These nutritional deficits exacerbate inflammation, impair immune defences and compromise organ function, increasing the risk of adverse clinical outcomes.8 9 Despite its recognised importance, accurately evaluating the nutritional status of critically ill patients with HF remains challenging, highlighting the demand for reliable, standardised assessment tools to identify at-risk individuals and guide interventions.
Various nutritional assessment tools have been developed, including the Controlling Nutritional Status (CONUT) and Prognostic Nutritional Index (PNI). The geriatric nutrition risk index (GNRI), initially developed to assess nutritional risk in elderly hospitalised patients, is now widely recognised as a practical, cost-effective tool for evaluating clinical outcomes in diverse patient populations.10 Among these tools, GNRI offers unique advantages through its straightforward, objective nutritional assessment that is simple to calculate and widely applicable in various clinical settings. Evidence strongly suggests that low GNRI values are associated with higher mortality in individuals with chronic illnesses,11,15 including HF.16 However, limited research has specifically explored its utility in assessing outcomes for critically ill patients with HF. Its role in combination with established models, such as the Sequential Organ Failure Assessment (SOFA) score, remains particularly under-investigated, especially in large cohort analyses.
This study aims to evaluate GNRI’s potential as a reliable tool for risk stratification and clinical decision-making in this vulnerable population by analysing a large cohort of ICU patients using robust statistical methods.
Materials and methods
Study design and setting
This retrospective cohort study utilised data from patients hospitalised for HF, retrieved from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database (version 3.1). The database contains critical care information for 76 540 patients admitted to ICUs at the Beth Israel Deaconess Medical Centre (Boston, MA, USA) between 2008 and 2022.17 The database is publicly accessible to researchers who have completed the Collaborative Institutional Training Initiative exam (Certification number 52 219 361 for Tang).
Following approval, the database was downloaded and stored locally. Structured Query Language (SQL)18 with PostgreSQL (version 13.0) and Navicat software (version 16.0) were used to identify the cohort and extract relevant clinical data. For clinical parameters with multiple recorded values during a single hospitalisation, only the initial value was included in the analysis. The data extraction code is publicly available on GitHub (https://github.com/MIT-LCP/mimic-iv). As this was a retrospective study using anonymised data from a public database, informed consent was waived.
Patients
Patients first admitted to ICU were identified from the MIMIC-IV database. Inclusion criteria required adult patients with HF, defined using the International Classification of Disease, Ninth or Tenth Revision (online supplemental table S1). Exclusion criteria included patients younger than 65 years, ICU stays shorter than 24 hours, or missing critical data, such as serum albumin, height and weight. After applying these criteria, 4058 patients with HF were included in the final analysis (figure 1).
Figure 1. The flowchart of patients’ selection. ICU, intensive care unit; MIMIC-IV, Medical Information Mart for Intensive Care IV.
Covariates
The following variables were included for all enrolled participants, based on published literature and clinical expertise: (1) Demographic characteristics: age, gender, race and body mass index (BMI), calculated as weight (kg)/height (m)². (2) Physical examination findings: heart rate and mean blood pressure (MBP). (3) Comorbidities: hypertension, diabetes, myocardial infarction, atrial fibrillation, stroke, chronic obstructive pulmonary disease (COPD), chronic kidney disease (CKD) and cancer. (4) Laboratory tests: haemoglobin, white blood cell count, platelet count, creatinine and albumin, measured within the first 24 hours of ICU admission. (5) Treatments: angiotensin-converting enzyme inhibitors/angiotensin receptor blockers (ACEI/ARB), beta-blockers, statins, diuretics, digoxin, vasoactive drugs (eg, dobutamine, dopamine, vasopressin, phenylephrine, norepinephrine, nitrates, nesiritide, epinephrine), haemodialysis and mechanical ventilation.
Additionally, the SOFA score during the first 24 hours of ICU admission was included as an important variable. The SOFA score is designed to objectively evaluate organ function across six systems: respiration, coagulation, liver, cardiovascular, central nervous system and renal. Scores for each domain range from 0 to 4, with higher scores indicating more severe organ dysfunction.
GNRI, the primary variable of interest, was calculated using the formula10: GNRI = (1.489×10 × serum albumin [g/dL]) + (41.7×weight (kg)/ideal body weight (kg)). The ideal body weight was calculated as follows: 0.75 × height (cm) - 62.5 for male patients, 0.60 × height (cm) - 40 for female patients. In addition, patients were categorised into two groups based on GNRI values10 19: those with no risk of malnutrition (GNRI >98) and those at risk of malnutrition (GNRI ≤98).
Outcomes
The primary outcomes of this study were in-hospital, 30-day and 1-year mortality following ICU admission. Mortality data were extracted from the MIMIC-IV database, which provides comprehensive survival information through linkage with the Social Security Death Master File and hospital records.
Statistical analysis
The normality of variable distributions was assessed using histograms, Q–Q plots and the Kolmogorov–Smirnov test. Continuous variables following a normal distribution were expressed as means with SD, whereas non-normally distributed variables were presented as medians and IQRs. Categorical variables were summarised as frequencies and percentages. Group comparisons for continuous variables were conducted using the independent samples Student’s t-test or the Mann–Whitney U test, depending on data distribution. Categorical data were compared using the chi-square test or Fisher’s exact test, as appropriate based on expected frequencies.
To examine the relationship between GNRI and in-hospital mortality, logistic regression models were employed to estimate ORs and 95% CIs. For 30-day and 1-year mortality analyses, Cox proportional hazards models were utilised, yielding HRs and corresponding 95% CIs. The proportional hazards assumption was validated using log–log plots, and interaction terms with survival time were included when necessary. Kaplan–Meier survival curves, stratified by malnutrition risk groups, were compared using the log-rank test.
Potential confounders were identified based on clinical relevance, prior studies, statistical significance in univariate analysis, or if their inclusion resulted in a greater than 10% change in effect size. Four analytical models were developed: Model 1 was unadjusted; Model 2 adjusted for demographic factors and vital signs; Model 3 further included SOFA scores, comorbidities and laboratory tests; Model 4, fully adjusted, incorporating treatment-related variables.
To evaluate non-linear dose-response relationships between GNRI and mortality, restricted cubic spline models were constructed with three knots located at the 25th, 50th and 75th percentiles of GNRI. Non-linearity was tested by introducing quadratic terms into regression models. In addition to categorical analysis, GNRI was analysed as a continuous variable using logistic regression for in-hospital mortality and Cox proportional hazards models for 30-day and 1- year mortality, complementing the restricted cubic spline analysis described above.
Receiver operating characteristic (ROC) curves were created for the GNRI+SOFA models and compared against ROC curves for the SOFA score alone to assess whether GNRI improved the predictive performance of these scoring systems. To evaluate the statistical significance of the differences in area under the curve (AUC) between the models, the DeLong test was applied. Additionally, the optimal cut-off points for each model were determined using the Youden index, which maximises the sum of sensitivity and specificity. The ROC curves were generated using bootstrap sampling with 500 iterations to estimate the variability of the AUC.
Missing data were addressed using multiple imputation with chained equations, following the method by Van Buuren and Groothuis-Oudshoorn (2011),20 implemented via the R package mice. This approach enhanced statistical power and minimised bias from missing data. The percentage of missing data for most variables was below 20%, as detailed in online supplemental table S2. Sensitivity analyses were conducted using complete-case data to validate the robustness of the primary findings, comparing effect sizes and p-values across models.
Subgroup analyses were conducted to explore the relationship between GNRI and mortality across predefined subgroup variables. To further test the robustness of the findings, additional association inference models were applied, including propensity score adjustment, propensity score matching (PSM) and inverse probability of treatment weighting (IPTW). PSM was performed using a 1:1 nearest-neighbour matching algorithm without replacement. Matching quality was assessed using standardised mean differences (SMD), with values <0.1 considered indicative of adequate balance between treatment groups. The propensity scores were estimated using identical covariates as those included in the multivariable regression models. Effect sizes and p-values obtained from these models were compared with evaluate consistency.
All statistical analyses were performed using R Statistical Software (version 4.2.2, available at http://www.R-project.org, The R Foundation) and the Free Statistics Analysis Platform (version 2.1, Beijing, China, http://www.clinicalscientists.cn/freestatistics). Statistical significance was set at a two-sided P-value below 0.05.
Results
Baseline characteristics
The clinical and demographic characteristics of the patients are presented in table 1, highlighting distinctions between those identified as having malnutrition risk and those without it. The average age of participants was 77.9 years, with males comprising a slight majority (55.3%). Individuals in the malnutrition risk category tended to be older; however, no notable differences were observed in terms of gender or racial composition. Patients classified as at risk for malnutrition exhibited a significantly lower BMI, indicative of potential nutritional deficiencies.
Table 1. Baseline characteristics of the study participants.
| Variables | Total (n=4 058) | Malnutrition risk | p Value | |
|---|---|---|---|---|
| With risk (n=2055) | Without risk (n=2 003) | |||
| Demographics | ||||
| Age (year) | 77.9±7.8 | 79.0±7.8 | 76.8±7.7 | < 0.001 |
| Gender male, n (%) | 2245 (55.3) | 1160 (56.4) | 1085 (54.2) | 0.144 |
| Race, n (%) | 0.431 | |||
| White | 2819 (69.5) | 1416 (68.9) | 1403 (70) | |
| Other | 1239 (30.5) | 639 (31.1) | 600 (30.0) | |
| BMI (kg/m2) | 28.1±6.0 | 24.3±3.9 | 32.0±5.1 | < 0.001 |
| Vital signs | ||||
| Heart rate (min−1) | 88.6±21.1 | 89.8±21.5 | 87.5±20.7 | < 0.001 |
| MBP, (mm Hg) | 80.9±18.0 | 79.6±17.9 | 82.1±18.1 | < 0.001 |
| SOFA | 7.0 (4.0, 10.0) | 7.0 (4.0, 10.0) | 6.0 (4.0, 9.0) | < 0.001 |
| Comorbidity, n (%) | ||||
| Hypertension | 3621 (89.2) | 1782 (86.7) | 1839 (91.8) | < 0.001 |
| Diabetes | 1748 (43.1) | 755 (36.7) | 993 (49.6) | < 0.001 |
| Myocardial infarction | 1505 (37.1) | 757 (36.8) | 748 (37.3) | 0.738 |
| Atrial fibrillation | 2646 (65.2) | 1295 (63) | 1351 (67.4) | 0.003 |
| Stroke | 620 (15.3) | 317 (15.4) | 303 (15.1) | 0.792 |
| COPD | 1513 (37.3) | 733 (35.7) | 780 (38.9) | 0.031 |
| CKD | 1866 (46.0) | 905 (44.0) | 961 (48.0) | 0.012 |
| Cancer | 552 (13.6) | 324 (15.8) | 228 (11.4) | < 0.001 |
| Laboratory tests | ||||
| Haemoglobin, (g/dL) | 10.2±2.2 | 9.8±2.1 | 10.6±2.3 | < 0.001 |
| WBC (K/μL) | 11.2 (8.1, 15.5) | 11.4 (8.1, 15.7) | 11.1 (8.1, 15.2) | 0.258 |
| Platelet (K/μL) | 191.0 (137.0, 259.0) | 190.0 (131.0, 263.0) | 192.0 (141.0, 254.0) | 0.399 |
| Creatinine (mg/dL) | 1.4 (1.0, 2.2) | 1.3 (0.9, 2.2) | 1.4 (1.0, 2.3) | < 0.001 |
| Albumin (g/dL) | 3.0±0.6 | 2.7±0.5 | 3.3±0.5 | < 0.001 |
| Treatments, n (%) | ||||
| ACEI/ARB | 749 (18.5) | 354 (17.2) | 395 (19.7) | 0.041 |
| Beta-blockers | 2801 (69.0) | 1412 (68.7) | 1389 (69.3) | 0.662 |
| Statin | 2491 (61.4) | 1205 (58.6) | 1286 (64.2) | < 0.001 |
| Diuretics | 3221 (79.4) | 1569 (76.4) | 1652 (82.5) | < 0.001 |
| Digoxin | 463 (11.4) | 257 (12.5) | 206 (10.3) | 0.026 |
| Vasoactive agents | 2971 (73.2) | 1558 (75.8) | 1413 (70.5) | < 0.001 |
| Haemodialysiss | 685 (16.9) | 333 (16.2) | 352 (17.6) | 0.244 |
| Mechanical ventilation | 3747 (92.3) | 1899 (92.4) | 1848 (92.3) | 0.86 |
| Length of hospital stay (days) | 11.7 (6.8, 19.6) | 11.9 (7.0, 20.3) | 11.1 (6.5, 18.7) | 0.004 |
| Mortality, n (%) | ||||
| In-hospital | 1099 (27.1) | 628 (30.6) | 471 (23.5) | < 0.001 |
| 30-day | 1331 (32.8) | 783 (38.1) | 548 (27.4) | < 0.001 |
| 1 year | 2165 (53.4) | 1242 (60.4) | 923 (46.1) | < 0.001 |
ACEI/ARB, angiotensin-converting enzyme inhibitors/angiotensin receptor blockers ; BMI, body mass index; CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; ICU, intensive care unit; MBP, mean blood pressure; SOFA, sequential organ failure assessment; WBC, white blood cell.
Patients in the malnutrition risk group displayed elevated SOFA scores and impaired haemodynamic parameters, such as increased heart rates and decreased MBP, both indicative of more severe illness. Although diabetes was less common among these individuals, conditions such as COPD, CKD and cancer occurred more frequently. Conversely, conditions such as hypertension and atrial fibrillation were less prevalent, suggesting distinct comorbidity profiles.
Laboratory findings revealed that patients in the malnutrition risk group had significantly reduced haemoglobin and albumin levels, indicating poorer overall health and nutritional status. Additionally, creatinine levels, though slightly lower, showed a consistent statistical difference.
Treatment strategies also varied between the groups, likely as a response to differing clinical needs. Those with malnutrition risk were less frequently prescribed ACEI/ARB and statins but were more often administered vasoactive agents and digoxin. Furthermore, these patients faced worse outcomes, as evidenced by higher rates of in-hospital, 30-day and 1 year mortality compared with individuals without malnutrition risk.
Association between GNRI and outcomes
In the Kaplan–Meier analysis, patients with malnutrition risk demonstrated significantly worse survival probabilities compared with those without risk for both 30-day (figure 2A) and 1-year (figure 2B) mortality (all p<0.001).
Figure 2. Kaplan–Meier survival curves for 30-day(A) and 1-year(B) mortality.
Patients with GNRI >98 demonstrated significantly lower risk of in-hospital mortality (OR=0.70, 95% CI 0.61 to 0.80) compared with those with GNRI ≤98 (reference group), indicating a protective effect of adequate nutritional status. After adjusting for confounding factors, including demographics, SOFA scores, comorbidities, laboratory tests and treatments in models 1, 2, 3 and 4, this relationship became less pronounced. However, GNRI ≤98 remained independently associated with a higher risk of adverse outcomes. Similarly, patients with GNRI >98 showed significantly lower risks of 30-day (HR=0.67, 95% CI 0.60 to 0.75) and 1 year (HR=0.67, 95% CI 0.61 to 0.73) mortality compared with those with GNRI ≤98. Adjustments for confounders did not eliminate this relationship, highlighting the strong association between malnutrition risk and mortality (table 2).
Table 2. Association between GNRI and mortality in heart failure patients.
| Models | Mortality | ||
|---|---|---|---|
| In-hospital OR (95% CI) | 30-day HR (95% CI) | 1-year HR (95% CI) | |
| Model 1 | 0.70 (0.61 to 0.80) | 0.67 (0.60 to 0.75) | 0.67 (0.61 to 0.73) |
| Model 2 | 0.74 (0.64 to 0.85) | 0.73 (0.65 to 0.81) | 0.74 (0.68 to 0.81) |
| Model 3 | 0.83 (0.71 to 0.97) | 0.80 (0.71 to 0.89) | 0.75 (0.69 to 0.83) |
| Model 4 | 0.82 (0.69 to 0.96) | 0.79 (0.70 to 0.88) | 0.76 (0.69 to 0.83) |
Note: Malnutrition with risk group as reference.
Model 1 adjust for: none.
Model 2 adjust for: demographic information (age, gender, race, BMI) and vital signs (heart rate, MBP).
Model 3 adjust for: Model 2+SOFA, comorbidities (hypertension, diabetes, myocardial infarction, atrial fibrillation, stroke, COPD, CKD, Cancer) and laboratory test (haemoglobin, WBC, platelet, creatinine, albumin).
Model 4 adjust for: Model 3+treatments (ACEI/ARB, beta blockers, statin, diuretics, digoxin, vasoactive agents, mechanical ventilation, haemodialysis).
ACEI/ARB, angiotensin-converting enzyme inhibitors/angiotensin receptor blockers; BMI, body mass index; CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; GNRI, geriatric nutritional risk index; MBP, mean blood pressure; SOFA, sequential organ failure assessment; WBC, white blood cell.
We utilised restricted cubic splines to explore the dose-response relationship between GNRI and mortality outcomes in patients with HF. As shown in figure 3, an L-shaped non-linear association was observed for in-hospital (A) and 1-year (C) mortality (P for non-linearity <0.05), while a linear relationship was identified for 30-day (B) mortality (P for non-linearity=0.055).
Figure 3. Dose-response relationship between the GNRI and in-hospital(A), 30-day(B), 1-year(C) mortality in patients with heart failure. Solid and dashed lines indicate the predicted value and 95% CI.
Online supplemental table S4 presents the linear association between GNRI as a continuous variable and mortality outcomes. In the fully adjusted model, each unit increase in GNRI was associated with an 8% reduction in in-hospital mortality risk (OR=0.92, 95% CI 0.87 to 0.97), 10% reduction in 30-day mortality risk (HR=0.90, 95% CI 0.87 to 0.94) and 11% reduction in 1-year mortality risk (HR=0.89, 95% CI 0.86 to 0.92), all statistically significant (p<0.001).
Adding GNRI to SOFA scores was significantly associated with enhanced predictive accuracy: in-hospital AUC=0.724 (95% CI 0.706 to 0.741), 30-day AUC=0.706 (95% CI 0.689 to 0.723) and 1-year AUC=0.670 (95% CI 0.654 to 0.687). (All p<0.05, as determined by the DeLong test (figure 4).
Figure 4. Incremental effect of the GNRI for predicting in-hospital(A), 30-day(B), and 1-year(C) mortality. AUC, area under the curve; GNRI, geriatric nutrition risk index; SOFA, sequential organ failure assessment.
Subgroup and additional analyses
Across subgroups stratified by age, gender, race, hypertension, myocardial infarction, stroke, renal failure and diabetes, no significant interactions were observed (all p>0.05). These findings suggest that the association between GNRI and mortality is consistent across diverse patient subgroups (online supplemental figures S1–S3).
To address potential confounding, PSM was applied to balance baseline characteristics. The SMD for all covariates after matching was less than 0.1, indicating adequate balance between groups (online supplemental table S3). Propensity score distribution analysis (online supplemental figure S4) demonstrated that both matching and weighting approaches effectively improved covariate balance between treatment groups. Online supplemental figure S5 presents SMD values for all covariates before and after matching, as well as after inverse probability weighting. Following both matching and weighting procedures, the majority of variables achieved SMD values <0.1, confirming adequate balance between groups. Even after applying PSM and inverse IPTW, GNRI ≤98 remained independently associated with an increased risk of mortality, further supporting the robustness of this relationship. Additionally, a multivariable regression analysis limited to patients with complete data yielded results that were consistent with the primary analysis, further validating the robustness of the association (table 3).
Table 3. Associations between GNRI and mortality in the sensitivity analyses.
| Analysis | In-hospital | 30-day | 1-year |
|---|---|---|---|
| OR (95% CI) | HR (95% CI) | HR (95% CI) | |
| Multivariable analysis | 0.82 (0.69 to 0.96) | 0.79 (0.70 to 0.88) | 0.76 (0.69 to 0.83) |
| After dropping missing data | 0.80 (0.68 to 0.95) | 0.77 (0.69 to 0.87) | 0.75 (0.68 to 0.83) |
| With matching | 0.82 (0.69 to 0.96) | 0.77 (0.68 to 0.88) | 0.75 (0.68 to 0.83) |
| With inverse probability weighting | 0.87 (0.76 to 1.00) | 0.82 (0.73 to 0.91) | 0.77 (0.71 to 0.84) |
| Adjusted for propensity score | 0.85 (0.73 to 0.99) | 0.80 (0.72 to 0.90) | 0.77 (0.70 to 0.84) |
Note: Multivariable Cox proportional hazards analysis adjusted for all covariates specified in Model four above. Malnutrition risk group served as reference category.
GNRI, geriatric nutritional risk index.
Discussion
This study highlights GNRI independently associated with mortality in critically ill patients with HF. GNRI ≤98 was strongly associated with higher in-hospital, 30-day and 1- year mortality, even after adjusting for confounders. An L-shaped relationship was observed for in-hospital and 1- year mortality, while a linear correlation emerged for 30-day mortality. The linear continuous variable analysis provides clinically interpretable estimates of the protective effect of GNRI. These findings suggest that nutritional status improvements, as reflected by GNRI increases, may be associated with incremental survival benefits across the entire range of values. This linear relationship, combined with our spline analysis showing threshold effects, offers clinicians both simplified prognostic tools and detailed insights into GNRI’s complex relationship with mortality. Incorporating GNRI into SOFA models was associated with enhanced predictive accuracy, supporting its potential role.
Our results align with earlier studies that recognised GNRI as a significant marker for mortality risk. For example, Yamada et al found that low GNRI values correlated with increased mortality in elderly patients suffering from chronic diseases,21 while Bouillanne et al confirmed GNRI’s reliability in evaluating malnutrition risks in hospitalised populations.10 In contrast to prior studies focusing on stable patients, our research highlights GNRI’s prognostic utility in critically ill patients with HF and its potential to improve SOFA score-based predictions. Moreover, the identified L-shaped relationship between GNRI and mortality adds new insights into its non-linear association with ICU outcomes.
Our results demonstrate that GNRI≤98 identifies high-risk patients with a 30% increase in in-hospital mortality odds and 33% increased hazard for both short-term and long-term mortality. This robust risk stratification, combined with GNRI’s simple calculation using readily available parameters (serum albumin and anthropometric data), supports its integration into routine ICU assessment. The L-shaped dose-response relationship reveals a critical threshold effect around GNRI=98, suggesting that nutritional interventions targeting patients near this threshold may yield maximum mortality reduction benefits. Furthermore, the enhanced predictive performance when combined with SOFA scores (AUC improvement from 0.718 to 0.724 for in-hospital mortality) demonstrates GNRI’s value as a complementary risk assessment tool that captures nutritional-metabolic dimensions not reflected in traditional organ dysfunction scoring.
These findings support a tiered approach: immediate comprehensive nutritional assessment for patients with GNRI ≤98, enhanced monitoring for those approaching this threshold (GNRI 98–105), and integration into multidisciplinary ICU protocols. The threshold effect observed supports early, aggressive nutritional intervention strategies rather than reactive approaches, potentially translating modest GNRI improvements into substantial mortality reduction.
The association between GNRI and mortality in critically ill patients with HF can be attributed to several mechanisms. Malnutrition, indicated by low GNRI, is associated with increased systemic inflammation, oxidative stress and immune dysfunction, corresponding to increased vulnerability.10 Low serum albumin, a key GNRI component, reflects both malnutrition and chronic inflammation, associated with worse prognosis.22 Additionally, muscle mass loss impairs cardiac and respiratory strength, hindering recovery.23 These factors underscore the importance of GNRI in risk stratification and outcome prediction.
Strengths and limitations
This study employed advanced methods to strengthen its findings. Restricted cubic spline analysis identified a novel L-shaped association between GNRI and in-hospital and 1 year mortality. Using robust methods like PSM and IPTW, the study minimised confounding and validated its findings using the MIMIC-IV database, offering insights into GNRI’s prognostic value.
However, this retrospective single-centre study has limitations. First, its retrospective design limits the ability to establish causality and generalisability. This single-centre ICU cohort may limit generalisability due to differences in patient characteristics across healthcare settings. However, consistent associations across subgroups support the internal validity of our findings, which are most applicable to similar ICU populations. Second, nutritional status was assessed exclusively through GNRI due to data availability constraints in the MIMIC-IV database. While alternative nutritional assessment tools such as CONUT and PNI would have provided valuable comparative insights, essential variables for these calculations (total cholesterol and lymphocyte counts) had prohibitively high missing rates (89.60% and 43.94%, respectively), limiting our analysis to only 6.2% of the total cohort if multiple indices were employed. Third, despite advanced statistical adjustments like PSM, residual confounding cannot be ruled out. Finally, our analysis was limited to baseline GNRI values and did not capture dynamic changes during hospitalisation, which may provide additional prognostic information. While our study demonstrated robust associations between GNRI and mortality, the absence of direct cardiac function metrics represents an important limitation. Although we included the cardiovascular component of the SOFA score, haemodynamic parameters and HF-specific treatments to indirectly reflect cardiac function, these proxies may not fully capture the complexity of cardiac dysfunction.
Future studies should validate these findings through multi-centre, prospective designs with comprehensive data collection protocols that enable systematic comparison of multiple nutritional assessment tools, include broader nutritional assessments incorporating dietary intake and body composition analysis, and examine temporal changes in GNRI during hospitalisation to optimise clinical decision-making and guide personalised interventions.
Conclusions
This study emphasises the role of GNRI as a valuable tool for improving risk stratification in critically ill HF patients with HF. Integrating GNRI into clinical practice could guide personalised interventions to address malnutrition. Future research should validate these findings in diverse cohorts, explore additional biomarkers and examine dynamic changes in GNRI to optimise care and improve outcomes.
Patient and public involvement
Patients and the public were not involved in the design, conduct, reporting or dissemination plans of this research. This retrospective cohort study utilised the publicly available, de-identified MIMIC-IV database, which contains historical clinical data without direct patient contact or recruitment. The nature of this secondary data analysis did not provide opportunities for patient and public involvement in research question formulation, outcome measure selection, study design, or interpretation of results.
Supplementary material
Acknowledgements
The present study data was based on the MIMIC-IV database. We would like to thank all staff and patients involved in the construction of the MIMIC-IV database.
Footnotes
Funding: This research was supported by the Tianjin Science and Technology Bureau - Research on Intelligent 3D Quantitative Analysis Technology of Coronary Bifurcation Lesions Using Optical Coherence Tomography (OCT) (Project No. 24YDTPJC00360).
Prepublication history and additional supplemental material for this paper are available online. To view these files, please visit the journal online (https://doi.org/10.1136/bmjopen-2025-100690 ).
Provenance and peer review: Not commissioned; externally peer reviewed.
Patient consent for publication: Not applicable.
Ethics approval: This study involves human participants, but the studies involving human participants were reviewed and approved by MIMIC-IV and eICU-CRD databases were approved by the institutional review boards of the Massachusetts Institute of Technology and Beth Israel Deaconess Medical Center. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements. Informed consent was obtained from all subjects and/or their legal guardian(s). Due to the retrospective nature of the study, the institutional review board waived the need for obtaining informed consent in the manuscript. exempted this study. Participants gave informed consent to participate in the study before taking part.
Data availability free text: Publicly available datasets were analysed in this study. This data can be found here: https://mimic.physionet.org/. Authorisation codes 52219361 for Tang facilitated access. Further requests can be directed to the corresponding author.
Patient and public involvement: Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Data availability statement
Data are available in a public, open access repository.
References
- 1.Savarese G, Becher PM, Lund LH, et al. Global burden of heart failure: a comprehensive and updated review of epidemiology. Cardiovasc Res. 2023;118:3272–87. doi: 10.1093/cvr/cvac013. [DOI] [PubMed] [Google Scholar]
- 2.McDonagh TA, Metra M, Adamo M, et al. 2023 Focused Update of the 2021 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure. Eur Heart J. 2023;44:3627–39. doi: 10.1093/eurheartj/ehad195. [DOI] [PubMed] [Google Scholar]
- 3.Valentová M, von Haehling S, Doehner W, et al. Liver dysfunction and its nutritional implications in heart failure. Nutrition. 2013;29:370–8. doi: 10.1016/j.nut.2012.06.002. [DOI] [PubMed] [Google Scholar]
- 4.Krack A, Sharma R, Figulla HR, et al. The importance of the gastrointestinal system in the pathogenesis of heart failure. Eur Heart J. 2005;26:2368–74. doi: 10.1093/eurheartj/ehi389. [DOI] [PubMed] [Google Scholar]
- 5.Kalantar-Zadeh K, Anker SD, Horwich TB, et al. Nutritional and anti-inflammatory interventions in chronic heart failure. Am J Cardiol. 2008;101:89E–103E. doi: 10.1016/j.amjcard.2008.03.007. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Poehlman ET, Scheffers J, Gottlieb SS, et al. Increased resting metabolic rate in patients with congestive heart failure. Ann Intern Med. 1994;121:860–2. doi: 10.7326/0003-4819-121-11-199412010-00006. [DOI] [PubMed] [Google Scholar]
- 7.Lv S, Ru S. The prevalence of malnutrition and its effects on the all-cause mortality among patients with heart failure: A systematic review and meta-analysis. PLoS ONE. 2021;16:e0259300. doi: 10.1371/journal.pone.0259300. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Lin H, Zhang H, Lin Z, et al. Review of nutritional screening and assessment tools and clinical outcomes in heart failure. Heart Fail Rev. 2016;21:549–65. doi: 10.1007/s10741-016-9540-0. [DOI] [PubMed] [Google Scholar]
- 9.Rahman A, Jafry S, Jeejeebhoy K, et al. Malnutrition and Cachexia in Heart Failure. J Parenter Enteral Nutr. 2016;40:475–86. doi: 10.1177/0148607114566854. [DOI] [PubMed] [Google Scholar]
- 10.Bouillanne O, Morineau G, Dupont C, et al. Geriatric Nutritional Risk Index: a new index for evaluating at-risk elderly medical patients. Am J Clin Nutr. 2005;82:777–83. doi: 10.1093/ajcn/82.4.777. [DOI] [PubMed] [Google Scholar]
- 11.Pressoir M, Desné S, Berchery D, et al. Prevalence, risk factors and clinical implications of malnutrition in French Comprehensive Cancer Centres. Br J Cancer. 2010;102:966–71. doi: 10.1038/sj.bjc.6605578. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Takahashi H, Ito Y, Ishii H, et al. Geriatric nutritional risk index accurately predicts cardiovascular mortality in incident hemodialysis patients. J Cardiol. 2014;64:32–6. doi: 10.1016/j.jjcc.2013.10.018. [DOI] [PubMed] [Google Scholar]
- 13.Beberashvili I, Azar A, Sinuani I, et al. Geriatric nutritional risk index, muscle function, quality of life and clinical outcome in hemodialysis patients. Clin Nutr. 2016;35:1522–9. doi: 10.1016/j.clnu.2016.04.010. [DOI] [PubMed] [Google Scholar]
- 14.Lee JS, Choi HS, Ko YG, et al. Performance of the Geriatric Nutritional Risk Index in predicting 28-day hospital mortality in older adult patients with sepsis. Clin Nutr. 2013;32:843–8. doi: 10.1016/j.clnu.2013.01.007. [DOI] [PubMed] [Google Scholar]
- 15.Wada H, Dohi T, Miyauchi K, et al. Prognostic Impact of the Geriatric Nutritional Risk Index on Long-Term Outcomes in Patients Who Underwent Percutaneous Coronary Intervention. Am J Cardiol. 2017;119:1740–5. doi: 10.1016/j.amjcard.2017.02.051. [DOI] [PubMed] [Google Scholar]
- 16.Li H, Cen K, Sun W, et al. Prognostic value of geriatric nutritional risk index in elderly patients with heart failure: a meta-analysis. Aging Clin Exp Res. 2021;33:1477–86. doi: 10.1007/s40520-020-01656-3. [DOI] [PubMed] [Google Scholar]
- 17.Johnson AEW, Bulgarelli L, Shen L, et al. MIMIC-IV, a freely accessible electronic health record dataset. Sci Data. 2023;10:1. doi: 10.1038/s41597-022-01899-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Johnson AEW, Stone DJ, Celi LA, et al. The MIMIC Code Repository: enabling reproducibility in critical care research. J Am Med Inform Assoc. 2018;25:32–9. doi: 10.1093/jamia/ocx084. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Zhao Y, Ge N, Xie D, et al. The geriatric nutrition risk index versus the mini-nutritional assessment short form in predicting postoperative delirium and hospital length of stay among older non-cardiac surgical patients: a prospective cohort study. BMC Geriatr. 2020;20:107. doi: 10.1186/s12877-020-1501-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Van Buuren S, Groothuis-Oudshoorn K. mice: Multivariate Imputation by Chained Equations in R. J Stat Softw. 2011;45:1–67. doi: 10.18637/jss.v045.i03. [DOI] [Google Scholar]
- 21.Yamada K, Furuya R, Takita T, et al. Simplified nutritional screening tools for patients on maintenance hemodialysis. Am J Clin Nutr. 2008;87:106–13. doi: 10.1093/ajcn/87.1.106. [DOI] [PubMed] [Google Scholar]
- 22.Don BR, Kaysen G. Serum albumin: relationship to inflammation and nutrition. Semin Dial. 2004;17:432–7. doi: 10.1111/j.0894-0959.2004.17603.x. [DOI] [PubMed] [Google Scholar]
- 23.Anker SD, Coats AJ. Cardiac cachexia: a syndrome with impaired survival and immune and neuroendocrine activation. Chest. 1999;115:836–47. doi: 10.1378/chest.115.3.836. [DOI] [PubMed] [Google Scholar]