Nutritional Score Improves ICU Risk Prediction After Major Abdominal Tumor Surgery

In a retrospective study of 1,195 adults undergoing hepatobiliary or pancreatic tumor surgery, a preoperative nutrition score called the geriatric nutritional risk index, or GNRI, improved prediction of postoperative intensive care unit admission more than any of five other nutrition- and inflammation-based indices examined. The clinical model had an area under the receiver operating characteristic curve, or AUC, of 0.766; adding GNRI raised the apparent AUC to 0.800. Internal validation reduced the GNRI-extended model’s AUC to 0.780, but the score still produced the largest observed improvement over the clinical model. Importantly, albumin alone achieved a similar apparent AUC of 0.799, suggesting that much of GNRI’s predictive information came from a routinely available laboratory measure rather than from the composite formula as a whole.

The analysis used the publicly accessible INSPIRE perioperative database, which contains anonymized records from patients who underwent surgery and anesthesia at Seoul National University Hospital between January 2011 and December 2020. Eligible participants were at least 18 years old, had a hepatobiliary or pancreatic tumor, underwent a qualifying operation, and had complete information for all six indices and the postoperative ICU outcome. The study population included 451 liver operations, 587 gallbladder operations and 157 pancreatic operations. Its mean age was 61.74 years, 69.0% of participants were male, and the median operative duration was 3.08 hours. Overall, 141 patients, or 11.8%, were admitted to the ICU after surgery.

Six indices derived from preoperative anthropometric and laboratory data were compared: GNRI, prognostic nutritional index, or PNI; hemoglobin, albumin, lymphocyte and platelet score, or HALP; leukocyte-to-albumin ratio, or LAR; systemic immune-inflammation index, or SII; and advanced lung cancer inflammation index, or ALI. Measurements collected within seven days before surgery were eligible, and the value closest to surgery was selected when more than one was available. Because the indices had different scales and distributions, researchers applied a logarithmic transformation and standardized each one before estimating its association with ICU admission. GNRI was also examined in prespecified categories below 82, from 82 to below 98, and at least 98; these cutoffs came from published nutritional-risk classifications and were not presented as optimal thresholds for this surgical population.

After adjustment for demographic characteristics, American Society of Anesthesiologists physical-status class, operation type, creatinine, chronic liver disease, diabetes, hypertension and major cardiovascular disease, all six indices were associated with postoperative ICU admission. A one-standard-deviation increase in transformed GNRI was associated with lower adjusted odds of ICU admission, with an odds ratio of 0.601 and a 95% confidence interval of 0.497 to 0.728. The same directional pattern appeared for PNI, HALP and ALI, whose adjusted odds ratios were 0.650, 0.664 and 0.674, respectively. Higher LAR and SII, which represent less favorable inflammatory profiles, were associated with greater odds: 1.434 for LAR and 1.430 for SII. These are observational associations and do not demonstrate that nutritional or inflammatory conditions caused the ICU admissions.

Category analyses showed graded relationships across all six measures. Compared with GNRI below 82, a value of at least 98 was associated with an adjusted odds ratio of 0.209 for ICU admission, with a 95% confidence interval of 0.080 to 0.548. Patients in the highest tertiles of PNI, HALP and ALI had lower adjusted odds than those in the lowest tertiles, with odds ratios of 0.335, 0.379 and 0.471. The highest tertiles of LAR and SII were associated with higher odds, with odds ratios of 2.060 and 2.393. Trend tests were statistically significant for every index, at p values no greater than 0.003. Restricted cubic spline analyses also found gradual dose-response relationships, but none showed a statistically significant nonlinear component.

For prediction, each index was added separately to the same adjusted clinical model. GNRI produced the largest observed rise in AUC, increasing it by 0.033 before rounding, from 0.766 to 0.800. The corresponding increments were 0.021 for PNI and LAR, 0.020 for HALP, 0.018 for ALI and 0.017 for SII. Adding GNRI reduced the Brier score from 0.094 to 0.092, while PNI produced the same apparent reduction. GNRI also had an integrated discrimination improvement of 0.028, with a 95% confidence interval of 0.013 to 0.045. Decision-curve analysis suggested potential net benefit over parts of the evaluated threshold range, but the findings describe model performance rather than evidence that using the scores would improve patient outcomes.

Bootstrap internal validation was used to estimate the optimism of models evaluated in the same cohort. The corrected AUC was 0.744 for the clinical model and 0.780 for the model that included GNRI, with a 95% confidence interval of 0.744 to 0.816. The mean optimism for the GNRI model was 0.020, and its mean bootstrap validation calibration slope was 0.897, indicating some degree of shrinkage. Corrected AUCs for models containing the other five indices ranged from 0.761 to 0.767. This procedure assessed internal validity only; it did not test the models in an independent hospital or patient population. A complete-case sensitivity analysis produced similar results, with the clinical AUC rising from 0.769 to 0.803 after GNRI was added.

Component models revealed that GNRI’s apparent advantage over the clinical model was largely shared by albumin. Albumin alone and albumin plus the capped actual-to-ideal weight ratio each yielded an AUC of 0.799, compared with 0.800 for GNRI. The weight ratio alone yielded an AUC of 0.768, and adding it to albumin did not materially improve discrimination. Once GNRI was included, adding PNI, HALP, SII, ALI or LAR changed the AUC by only −0.001 to 0.006. Corresponding confidence intervals for continuous net reclassification improvement included zero, leaving any additional reclassification gains uncertain. These comparisons indicate that combining several nutrition- and inflammation-related variables did not provide much information beyond what was already available through albumin and standard clinical variables.

The findings also need to be interpreted in light of several limitations. INSPIRE contains data from one center, and ICU admission practices, monitoring strategies and resource availability may differ elsewhere. The recorded outcome did not distinguish planned from unplanned admission, so it may reflect institutional monitoring and allocation practices as well as postoperative illness severity. Important potential confounders—including tumor stage, functional status, frailty, sarcopenia, preoperative nutritional interventions and reliable measures of surgical extent—were unavailable or incomplete. Operative duration was known only after surgery and was therefore used only in sensitivity analysis. The indices also relied on single measurements that could be affected by transient inflammation, metabolism or hydration, and conventional GNRI thresholds may not be optimal for a surgical population of mixed ages.

Exploratory subgroup tests suggested possible interactions of GNRI with age and chronic liver disease, and of LAR with chronic liver disease. However, none of the 54 interaction tests remained significant after false-discovery-rate correction, so the study did not establish that the associations differed reliably across those groups. The results instead identify a promising line of perioperative research: simple preoperative markers may help estimate the probability of postoperative ICU use after complex abdominal tumor operations, but the apparent advantage of a composite nutrition score over albumin alone remains uncertain. Independent external validation, assessment of calibration in different clinical settings and evaluation of clinical impact would be needed before these measures could support decisions about monitoring, ICU allocation or nutritional care.

Subject of Research: Preoperative nutritional and inflammatory indices as predictors of postoperative ICU admission after hepatobiliary and pancreatic tumor surgery

Article Title: Association and incremental predictive performance of preoperative nutritional and inflammatory indices for postoperative ICU admission after hepatobiliary and pancreatic tumor surgery

Article References: Ou, L., & Wang, H. (2026). Association and incremental predictive performance of preoperative nutritional and inflammatory indices for postoperative ICU admission after hepatobiliary and pancreatic tumor surgery. Frontiers in Nutrition, 13, Article 1978034. https://doi.org/10.3389/fnut.2026.1978034

Image Credits: AI Generated

DOI: 10.3389/fnut.2026.1978034

Keywords: GNRI, postoperative ICU admission, nutritional risk, systemic inflammation, hepatobiliary surgery, pancreatic surgery, albumin, clinical prediction, perioperative risk, Association, incremental, predictive