Prognostic Value of Perioperative Inflammatory–Immune–Nutritional Indicators in Advanced Gallbladder Cancer

ABSTRACT

Objective

To explore the perioperative changes in inflammatory–immune–nutritional indicators in patients with advanced gallbladder cancer (GBC), defined as American Joint Committee on Cancer (AJCC) Stage III–IV, and to assess whether individual indicators provide incremental prognostic information beyond conventional clinicopathological factors.

Methods

Patients with pathologically confirmed advanced GBC who underwent surgery at the Department of Hepatobiliary Surgery, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, between 2010 and 2022 were retrospectively enrolled. Preoperative and postoperative Day 7 values of the albumin‐to‐alkaline phosphatase ratio (AAPR), albumin‐to‐globulin ratio (AGR), albumin–bilirubin score (ALBI), advanced lung cancer inflammation index (ALI), monocyte‐to‐lymphocyte ratio (MLR), neutrophil‐to‐lymphocyte ratio (NLR), prognostic nutritional index (PNI), and systemic inflammation response index (SIRI) were calculated. Perioperative dynamic changes were compared, and Cox proportional hazards models were constructed to evaluate the incremental prognostic value of these indicators.

Results

Among 103 patients with advanced GBC, all eight inflammatory–immune–nutritional indicators showed significant changes between the preoperative period and postoperative Day 7. Several preoperative and postoperative indicators were associated with cancer‐specific survival (CSS) in cutoff‐based exploratory analyses. After the incorporation of these indicators into clinicopathological models, most extended models showed varying degrees of improvement in discrimination and goodness of fit.

Conclusion

In this single‐center cohort, perioperative inflammatory–immune–nutritional indicators exhibit significant dynamic changes in advanced GBC and can improve, to a certain extent, the prognostic performance of models based on conventional clinicopathological characteristics. These indicators may serve as useful supplementary tools for risk stratification and individualized prognostic assessment in patients with advanced GBC.

Abbreviations

AAPR

albumin‐to‐alkaline phosphatase ratio

AGR

albumin‐to‐globulin ratio

AIC

Akaike information criterion

AJCC

American Joint Committee on Cancer

ALBI

albumin–bilirubin score

ALI

advanced lung cancer inflammation index

AUC

area under the curve

CI

confidence interval

C‐index

concordance index

CSS

cancer‐specific survival

DCA

Decision curve analysis

GBC

gallbladder cancer

HR

hazard ratio

MLR

monocyte‐to‐lymphocyte ratio

NLR

neutrophil‐to‐lymphocyte ratio

PNI

prognostic nutritional index

SIRI

systemic inflammation response index

1. Introduction

Gallbladder cancer (GBC) is the most common malignancy of the biliary tract [1]. It ranks 23rd among all cancers and sixth among digestive system malignancies, and is characterized by an insidious onset, rapid progression, aggressive biological behavior, and extremely poor prognosis, with an overall 5‐year survival rate of less than 20% [1, 2, 3, 4, 5, 6, 7]. With population growth, aging, and persistent regional differences in the distribution of risk factors, the numbers of incident cases, deaths, and disability‐adjusted life years continue to represent a substantial public health burden [8]. GBC is also one of the few malignancies with a pronounced sex difference; its incidence in women is nearly three times that in men, making it one of the uncommon female‐predominant digestive system cancers [9, 10, 11]. Because early‐stage disease lacks specific clinical manifestations, most patients are diagnosed at a locally advanced or metastatic stage, and only approximately 20%–25% are considered resectable at presentation [7, 10, 12, 13]. Although surgical techniques, perioperative management, and systemic therapies have improved in recent years, outcomes for GBC, particularly advanced GBC, remain unsatisfactory [3, 14]. Improving prognosis therefore remains a major clinical challenge.

Tumor stage is a key determinant of treatment strategy and prognosis in GBC. The eighth edition of the American Joint Committee on Cancer (AJCC) staging system is currently one of the most widely used staging frameworks for malignancies [15]. Compared with patients with early‐stage disease (AJCC Stages I and II), patients with advanced GBC (AJCC Stages III and IV) usually have deeper local invasion, more extensive lymph node involvement, or distant metastasis, and their survival outcomes are substantially worse [2, 16]. This population represents the most challenging subgroup and is therefore of particular research importance.

Conventional clinicopathological factors reflect the biological behavior of the tumor but provide limited information regarding the host condition. Tumor‐associated inflammation, immune microenvironment remodeling, and nutritional–metabolic imbalance have been shown to be closely associated with tumor development, treatment response, and survival outcomes [17]. Therefore, identifying readily available surrogate markers that complement conventional clinicopathological information is important for advanced GBC. In recent years, inflammatory–immune–nutritional indicators derived from routine blood and biochemical tests—including AAPR, AGR, ALBI, ALI, MLR, NLR, PNI, and SIRI—have become research hotspots in oncology because they are easily obtained, inexpensive, and dynamically measurable [18, 19, 20, 21, 22, 23, 24]. These indicators reflect inflammation, immune responsiveness, and nutritional reserve from different perspectives and have shown prognostic value in several digestive system malignancies. However, in advanced GBC, existing studies have mainly focused on a single indicator at a single time point. Systematic evidence regarding preoperative‐to‐postoperative dynamic changes, comparisons among different indicators, and their incremental value when incorporated into clinicopathological models remains limited [25, 26, 27, 28, 29]. In particular, whether indicators measured at postoperative Day 7, a clinically meaningful time point close to routine discharge and planning of subsequent adjuvant therapy, can improve survival prediction warrants further investigation.

Accordingly, conventional clinicopathological factors remain insufficient for capturing the inflammatory, immune, and nutritional status of the host. Based on a clinical cohort from the Department of Hepatobiliary Surgery of the National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College (hereafter referred to as our center), this study combined perioperative dynamic inflammatory–immune–nutritional indicators to construct and compare prognostic prediction models. We aimed to explore the potential value of these indicators in risk stratification and individualized prognostic assessment for advanced GBC, thereby improving understanding of disease characteristics and enhancing model performance.

2. Materials and Methods

2.1. Study Population

Data were retrospectively collected from patients with surgically confirmed GBC treated at our center between 2010 and 2022. Follow‐up was conducted until July 30, 2023. The inclusion criteria were as follows: (1) surgical treatment was performed; and (2) postoperative pathology confirmed gallbladder adenocarcinoma. The exclusion criteria were as follows: (1) death or loss to follow‐up within 1 month after surgery; (2) age < 18 years; and (3) severe comorbidities, including severe hypertension, diabetes mellitus, coronary heart disease, heart failure, or hepatic or renal failure [30]. After screening for patients with AJCC Stage III or IV disease, 103 eligible patients with advanced GBC were included.

2.2. Variables and Definitions

Demographic, pathological, treatment‐related, laboratory, and survival data were extracted from the electronic medical record system of our center. Demographic variables included age and sex. Pathological variables included maximum tumor diameter, tumor differentiation, T stage, N stage, M stage, and AJCC stage. Treatment‐related variables included surgical category, chemotherapy, and radiotherapy. The inflammatory–immune–nutritional indicators AAPR, AGR, ALBI, ALI, MLR, NLR, PNI, and SIRI were assessed both preoperatively and on postoperative Day 7.

Patients who did not undergo lymph node dissection or adjacent liver resection, or who did not undergo resection of metastatic lesions, were classified as having undergone non‐radical surgery. Chemotherapy and radiotherapy were recorded as binary variables according to whether receipt of the corresponding treatment was documented in the medical records.

Laboratory variables with missing values of < 20% were imputed using the missForest method, whereas variables that did not meet this criterion were excluded. Cancer‐specific survival (CSS) was defined as the interval from the date of surgery to death attributable to GBC. Death from GBC was coded as the event. Patients who remained alive at the end of follow‐up or were lost to follow‐up were censored at the date of their last known follow‐up. The inflammatory–immune–nutritional indicators were calculated from perioperative laboratory parameters as follows:

$$\text{AAPR}=\text{albumin}\left(\mathrm{g}/\mathrm{L}\right)/\text{alkaline phosphatase}\left(\mathrm{U}/\mathrm{L}\right)$$

$$\mathrm{AGR}=\text{albumin}\left(\mathrm{g}/\mathrm{L}\right)/\text{globulin}\left(\mathrm{g}/\mathrm{L}\right)$$

$$\text{ALBI}=0.66\times \log 10\left(\text{total bilirubin}\left(\text{umol}/\mathrm{L}\right)\right)-0.085\times \text{albumin}\left(\mathrm{g}/\mathrm{L}\right)$$

$$\begin{array}{c}\mathrm{ALI}=\mathrm{BMI}\left(\mathrm{kg}/{\text{m}}^2\right)\times \text{albumin}\left(\text{g}/\text{L}\right)/\mathrm{NLR}\hfill \end{array}$$

$$\begin{array}{c}\mathrm{MLR}=\text{monocyte count}\left(\times {10}^9\right)/\text{lymphocyte count}\left(\times {10}^9\right)\hfill \end{array}$$

$$\begin{array}{c}\text{SIRI}=\text{neutrophil count}\left(\times {10}^9\right)\times \text{monocyte count}\left(\times {10}^9\right)/\hfill \\ \text{lymphocyte count}\left(\times {10}^9\right)\hfill \end{array}$$

2.3. Statistical Analysis

Clinical and pathological data from the 103 patients included in this study were systematically organized and analyzed. Continuous variables are presented as medians with interquartile ranges, M (P25, P75), and categorical variables are presented as counts and percentages, n (%). Spearman correlation analysis was used to evaluate correlations among covariates. The paired Wilcoxon signed‐rank test was used to compare preoperative and postoperative day 7 inflammatory–immune–nutritional indicators. Continuous variables were visualized using combined box and scatter plots, with paired line plots illustrating individual level changes.

To assess the prognostic stratification ability of each inflammatory–immune–nutritional indicator, optimal cutoff values were determined in the full original cohort using maximally selected rank statistics, with the minimum proportion of observations in either resulting group restricted to 10%. Patients were subsequently categorized into low and high level groups according to the identified cutoff values, with the low‐level group serving as the reference category in the Cox regression analyses. Kaplan–Meier curves for CSS were plotted, and differences between groups were compared using the log‐rank test. A two‐sided p value < 0.05 was considered statistically significant.

Internal validation was performed using 1000 bootstrap resamples. The complete cutoff‐selection procedure was repeated within each bootstrap sample. The original full‐cohort cutoff values were retained for the primary exploratory analyses; bootstrap‐derived values were used to assess the internal stability and sampling variability of the selected thresholds rather than to define replacement clinical cutoffs.

To evaluate the incremental value of inflammatory–immune–nutritional indicators for survival prediction, multivariable Cox proportional hazards regression models were constructed. A baseline model (Model 1) was built using clinical and pathological variables, and extended models were constructed by adding indicators measured at different time points. Model 1 included age, sex, surgical radicality, maximum tumor diameter, differentiation, tumor stage, chemotherapy, and radiotherapy. Model 2 added the preoperative indicator to Model 1. Model 3 added the postoperative Day 7 indicator to Model 1. Model 4 added both the preoperative and postoperative Day 7 indicators to Model 1. Each inflammatory–immune–nutritional indicator was evaluated in a separate set of Cox models. Specifically, different indicators were not entered simultaneously into the same regression model. Harrell's concordance index (C‐index) was used to assess model discrimination, the Akaike information criterion (AIC) was used to assess goodness of fit, and time‐dependent areas under the receiver operating characteristic curve (AUCs) were calculated to evaluate predictive performance at 0.5, 1, 1.5, and 2 years. Higher C‐index and AUC values indicated stronger discriminative ability, whereas lower AIC values indicated better model fit. Calibration analysis was performed to assess the agreement between model‐predicted and observed CSS probabilities, with the diagonal reference line representing perfect calibration. Decision curve analysis (DCA) was additionally conducted to evaluate the potential clinical utility of Models 1–4 by comparing their net benefits across a range of threshold probabilities with the reference strategies of treating all patients and treating no patients; a higher net benefit within a given threshold range was considered indicative of greater potential clinical utility.

To further assess the robustness and potential heterogeneity of the findings, additional sensitivity and subgroup analyses were performed. First, the main analyses were repeated separately in patients with Stage III and Stage IV GBC to explore stage‐specific differences. In these stage‐stratified analyses, the cutoff values derived from the overall cohort were retained to ensure consistency and to avoid deriving potentially unstable new thresholds from the smaller stage‐specific subgroups. Second, a complete‐case sensitivity analysis was conducted using only patients with complete data and without any imputation. Because this complete‐case cohort represented a distinct analytical dataset, the optimal cutoff values were re‐derived within the complete‐case cohort using the same cutoff‐selection procedure as in the primary analysis.

All statistical analyses were performed using the R software (version 4.4.1).

3. Results

3.1. Prognostic Value of Inflammatory–Immune–Nutritional Indicators in Advanced GBC

3.1.1. Clinical and Pathological Characteristics

A total of 103 patients from our center were included. The median age was 62 years, and women accounted for 59.2% of the cohort.

The median maximum tumor diameter was 4.00 cm. Poor differentiation was predominant (69.9%). Pathological staging showed 50 patients (48.5%) with Stage III disease and 53 patients (51.5%) with Stage IV disease.

75 patients (72.8%) underwent radical surgery and 28 (27.2%) underwent non‐radical surgery. Among adjuvant treatments, 61 patients (59.2%) received chemotherapy and 10 (9.7%) received radiotherapy. During follow‐up, 47 of the 103 patients experienced cancer‐specific death.

Regarding laboratory parameters, the median preoperative carcinoembryonic antigen (CEA) level was 3.67 ng/mL, and the median preoperative carbohydrate antigen 19‐9 (CA19‐9) level was 81.87 U/mL. Preoperative inflammatory–immune–nutritional indicators were as follows: AAPR, 0.51 (0.37, 0.61); AGR, 1.50 (1.29, 1.67); ALBI, −2.86 (−3.09, −2.57); ALI, 466.89 (306.44, 665.19); MLR, 0.26 (0.18, 0.34); NLR, 2.10 (1.64, 3.11); PNI, 50.60 (47.45, 53.25); and SIRI, 0.94 (0.53, 1.63). Postoperative indicators were as follows: AAPR, 0.29 (0.21, 0.42); AGR, 1.36 (1.18, 1.52); ALBI, −2.01 (−2.23, −1.91); ALI, 180.44 (127.85, 227.54); MLR, 0.50 (0.39, 0.61); NLR, 4.59 (3.52, 6.20); PNI, 40.48 (37.85, 43.07); and SIRI, 3.36 (2.12, 4.73) (Table 1).

TABLE 1.

Clinical and pathological characteristics of the institutional cohort.

Variable	Distribution
n	103
Age, years	62.00 [56.50, 67.00]
Sex: male	42 (40.8%)
Sex: female	61 (59.2%)
Maximum tumor diameter, cm	4.00 [3.00, 5.50]
Differentiation: poorly differentiated	72 (69.9%)
Differentiation: moderately/well differentiated	31 (30.1%)
T stage: 1	1 (1.0%)
T stage: 2	22 (21.4%)
T stage: 3	61 (59.2%)
T stage: 4	19 (18.4%)
N stage: 0	35 (34.0%)
N stage: 1	39 (37.9%)
N stage: 2	29 (28.2%)
M stage: 0	72 (69.9%)
M stage: 1	31 (30.1%)
AJCC Stage: III	50 (48.5%)
AJCC Stage: IV	53 (51.5%)
Surgical category: radical surgery	75 (72.8%)
Surgical category: non‐radical surgery	28 (27.2%)
Chemotherapy: no	42 (40.8%)
Chemotherapy: yes	61 (59.2%)
Radiotherapy: no	93 (90.3%)
Radiotherapy: yes	10 (9.7%)
Preoperative AAPR	0.51 [0.37, 0.61]
Preoperative AGR	1.50 [1.29, 1.67]
Preoperative ALBI	−2.86 [−3.09, −2.57]
Preoperative ALI	466.89 [306.44, 665.19]
Preoperative MLR	0.26 [0.18, 0.34]
Preoperative NLR	2.10 [1.64, 3.11]
Preoperative PNI	50.60 [47.45, 53.25]
Preoperative SIRI	0.94 [0.53, 1.63]
Postoperative AAPR	0.29 [0.21, 0.42]
Postoperative AGR	1.36 [1.18, 1.52]
Postoperative ALBI	−2.01 [−2.23, −1.91]
Postoperative ALI	180.44 [127.85, 227.54]
Postoperative MLR	0.50 [0.39, 0.61]
Postoperative NLR	4.59 [3.52, 6.20]
Postoperative PNI	40.48 [37.85, 43.07]
Postoperative SIRI	3.36 [2.12, 4.73]

Correlation analysis of the included inflammatory–immune–nutritional indicators (preoperative and postoperative AAPR, AGR, ALBI, ALI, MLR, NLR, PNI, and SIRI) showed varying degrees of correlation among indicators. Some pairs, such as preoperative ALI and preoperative NLR, had relatively high absolute correlation coefficients, suggesting a potential risk of collinearity (Figure S1). However, in the present study, each indicator was incorporated independently into subsequent analyses rather than simultaneously entering the same regression model; therefore, correlations among indicators did not compromise the reliability of the subsequent analyses.

Paired analyses showed that all eight inflammatory–immune–nutritional indicators differed significantly between the preoperative period and postoperative Day 7 (p < 0.001). AAPR, AGR, ALI, and PNI decreased markedly after surgery, suggesting a perioperative decline in nutritional reserve and immune–nutritional status. ALBI increased significantly after surgery, indicating short‐term postoperative worsening of liver function‐related status. MLR, NLR, and SIRI increased significantly after surgery, reflecting enhanced systemic inflammation and disruption of inflammatory–immune homeostasis. Individual trend plots showed a generally consistent direction of change despite heterogeneity in the magnitude of change across patients. These findings suggest that surgical stress and perioperative host responses may lead to short‐term inflammatory activation, accompanied by deterioration in nutritional status and liver function‐related indicators (Figure 1).

FIGURE 1.

Preoperative and postoperative comparisons of the eight inflammatory–immune–nutritional indicators in the institutional cohort. (A) AAPR; (B) AGR; (C) ALBI; (D) ALI; (E) MLR; (F) NLR; (G) PNI; (H) SIRI.

3.2. Survival Impact of Inflammatory–Immune–Nutritional Indicators

After dichotomization of each inflammatory–immune–nutritional indicator based on its cutoff value, univariable Cox survival analysis showed that preoperative MLR, PNI, and SIRI were significantly associated with patient prognosis. Specifically, patients in the high‐MLR group had a significantly increased risk of death (HR = 2.580, 95% CI: 1.379–4.828, p = 0.003). PNI and SIRI groups also effectively distinguished survival risk (PNI: HR = 1.904, 95% CI: 1.031–3.519, p = 0.040; SIRI: HR = 2.041, 95% CI: 1.037–4.020, p = 0.039). Although the remaining preoperative indicators did not reach statistical significance, some showed a trend toward risk stratification (Figure 2, Table S1).

FIGURE 2.

Cancer‐specific survival curves after dichotomization of the eight preoperative inflammatory–immune–nutritional indicators in the institutional cohort. (A) AAPR; (B) AGR; (C) ALBI; (D) ALI; (E) MLR; (F) NLR; (G) PNI; (H) SIRI.

Among postoperative indicators, ALI, NLR, and PNI were significantly associated with prognosis. ALI showed relatively strong discriminative performance (HR = 2.745, 95% CI: 1.375–5.479, p = 0.004), and PNI and NLR also demonstrated significant prognostic value after grouping (PNI: HR = 2.320, 95% CI: 1.226–4.391, p = 0.010; NLR: HR = 0.383, 95% CI: 0.177–0.827, p = 0.015) (Figure 3, Table S1).

FIGURE 3.

Cancer‐specific survival curves after dichotomization of the eight postoperative inflammatory–immune–nutritional indicators in the institutional cohort. (A) AAPR; (B) AGR; (C) ALBI; (D) ALI; (E) MLR; (F) NLR; (G) PNI; (H) SIRI.

The cutoff derived from the complete original cohort, the bootstrap median and interquartile range, the 2.5th–97.5th percentile range of the bootstrap‐derived cutoffs, and the apparent and optimism‐corrected C‐indexes are presented in Table S2. The optimism‐corrected C‐indexes were consistently lower than the corresponding apparent C‐indexes, indicating varying degrees of optimism in the original cutoff‐based survival discrimination. The stability of the selected thresholds differed across indicators. Several postoperative indicators demonstrated relatively stable central estimates. For postoperative ALBI, PNI, and SIRI, the original cutoff values were identical to the corresponding bootstrap medians. The original and bootstrap‐derived central estimates for postoperative AAPR, AGR, and ALI were also relatively close. In contrast, wider bootstrap distributions were observed for some other indicators, indicating greater sensitivity to sampling variation.

Overall, cutoff‐based stratification suggested that inflammatory–immune–nutritional indicators, whether measured preoperatively or postoperatively, have prognostic stratification ability and can be used to distinguish high‐risk from low‐risk patients. These findings indicate that dynamic postoperative monitoring may provide additional information for survival prediction.

3.3. Influence of Inflammatory–Immune–Nutritional Indicators on Survival Models

The baseline model (Model 1) had a C‐index of 0.739 and an AIC of 369.986. Across the eight indicators, adding inflammatory–immune–nutritional indicators to the baseline model generally improved predictive performance to varying degrees, and models that simultaneously included preoperative and postoperative day 7 indicators (Model 4) were most often optimal or close to optimal. In the AAPR models, overall improvement was limited; Model 3 showed a slightly higher C‐index than the baseline model (0.741), but the AIC increased, suggesting limited incremental value. In the AGR models, Model 4 had the highest C‐index (0.754), the AIC decreased to 368.907, and the 2‐year AUC increased to 80.14%, suggesting modest improvement after dynamic incorporation of AGR. The ALBI models showed the most prominent and stable performance; Model 4 achieved a C‐index of 0.774, an AIC of 367.993, and AUCs of 89.23%, 87.67%, 84.97%, and 81.29% at 0.5, 1, 1.5, and 2 years, respectively, all superior to those of the baseline model.

In the ALI models, Model 4 achieved the highest C‐index (0.760), and the 1.5‐ and 2‐year AUCs reached 83.93% and 83.99%, respectively, indicating relatively high value for mid‐ and long‐term prognostic assessment when dynamic ALI was incorporated. In the MLR models, Model 4 had the highest C‐index (0.765) and the lowest AIC (367.310), and the 2‐year AUC increased to 81.64%, indicating improved overall predictive performance. In the NLR models, Model 4 also performed well, with the C‐index increasing to 0.760 and the AIC decreasing to 364.582; AUCs at all time points were generally higher than those of the baseline model, suggesting favorable predictive gain when preoperative and postoperative Day 7 NLR were jointly included. In the PNI models, Model 4 showed the best goodness of fit, with the AIC decreasing to 355.974, the lowest among all models; however, the C‐index increased only to 0.748, indicating a more pronounced improvement in model fit than in discrimination. By contrast, SIRI yielded only modest improvement; the C‐index of Model 4 increased to 0.748, and the 1‐ and 2‐year AUCs reached 87.28% and 78.13%, respectively, but the overall magnitude of improvement was limited.

Calibration and DCA were performed for Models 1–4 at 1 and 2 years (Figure S2). Overall, calibration was relatively better at 1 year, whereas greater deviations between predicted and observed CSS probabilities were observed at 2 years, with a tendency toward overestimation of survival in some low‐to‐intermediate predicted‐survival ranges. DCA showed that several extended models provided greater net benefit than the clinicopathological baseline model across selected threshold‐probability ranges; however, the magnitude and range of benefit varied across indicators, models, and prediction horizons.

Taken together, most indicators improved model performance to some extent after incorporation, and models that jointly included preoperative and postoperative Day 7 inflammatory–immune–nutritional indicators generally performed better or close to the best (Table 2).

TABLE 2.

Model comparison after combining the eight inflammatory–immune–nutritional indicators with clinical and pathological variables.

Model	C‐index	AIC	AUC: 0.5 y	AUC: 1 y	AUC: 1.5 y	AUC: 2 y
Model 1	0.739	369.986	87.02%	83.63%	80.02%	77.99%
AAPR Model 2	0.727	369.556	86.10%	81.99%	77.32%	74.89%
AAPR Model 3	0.741	370.717	87.24%	84.71%	80.84%	78.01%
AAPR Model 4	0.734	369.933	85.69%	82.59%	78.41%	74.83%
AGR Model 2	0.749	368.170	88.83%	82.14%	79.44%	78.52%
AGR Model 3	0.750	371.209	87.33%	84.98%	81.25%	79.12%
AGR Model 4	0.754	368.907	88.72%	82.59%	80.36%	80.14%
ALBI Model 2	0.757	369.873	90.49%	85.77%	81.43%	79.31%
ALBI Model 3	0.744	368.811	84.82%	83.32%	82.43%	79.36%
ALBI Model 4	0.774	367.993	89.23%	87.67%	84.97%	81.29%
ALI Model 2	0.739	371.536	88.39%	83.99%	79.35%	77.44%
ALI Model 3	0.756	362.011	84.97%	82.21%	83.71%	83.42%
ALI Model 4	0.760	362.375	85.27%	82.12%	83.93%	83.99%
MLR Model 2	0.756	367.632	88.18%	84.63%	81.81%	78.58%
MLR Model 3	0.744	370.258	87.95%	83.18%	81.14%	81.10%
MLR Model 4	0.765	367.310	89.21%	84.15%	81.78%	81.64%
NLR Model 2	0.742	370.745	89.56%	83.40%	81.86%	77.47%
NLR Model 3	0.741	365.429	86.03%	83.27%	79.11%	79.46%
NLR Model 4	0.760	364.582	88.56%	85.09%	82.59%	81.16%
PNI Model 2	0.734	361.537	86.87%	80.10%	77.63%	78.72%
PNI Model 3	0.744	365.304	80.31%	81.53%	82.53%	82.13%
PNI Model 4	0.748	355.974	80.68%	79.85%	80.69%	82.71%
SIRI Model 2	0.746	371.065	87.87%	84.83%	80.35%	78.32%
SIRI Model 3	0.744	369.176	87.35%	86.16%	79.00%	77.76%
SIRI Model 4	0.748	370.993	87.67%	87.28%	79.72%	78.13%

3.4. Sensitivity and Subgroup Analyses

Additional stage‐stratified and sensitivity analyses were performed to assess the robustness of the primary findings. In analyses conducted separately among patients with Stage III and Stage IV GBC, the overall patterns of prognostic associations and model‐performance changes were broadly consistent with those observed in the overall cohort, although some variability was observed across individual indicators (Figure S3–S8, Tables S3, S4). Similarly, the complete‐case sensitivity analysis, which excluded patients with imputed values, showed generally comparable trends to the primary analysis (Figures S9–S11, Table S5). Given the reduced sample sizes and number of events in these additional analyses, the results were considered exploratory and were not used to draw definitive subgroup‐specific conclusions.

4. Discussion

GBC is one of the most aggressive malignancies of the digestive system. Because early symptoms are obscure, most patients are diagnosed at an advanced stage, and improvement in population‐level survival remains limited [6, 10, 12, 31]. Therefore, beyond conventional clinicopathological factors, identifying convenient indicators that reflect systemic inflammation, immune status, and nutritional reserve is important for improving risk stratification and individualized prognostic assessment in patients with advanced GBC.

Advanced GBC is characterized by more pronounced enrichment of inflammation, immune regulation, cell proliferation, and related pathways [32, 33]. This suggests that during progression to advanced disease, GBC may undergo a shift from relatively stable structural and metabolic features toward inflammatory activation, immune remodeling, and a proliferative state. Preoperative changes in these indicators do not necessarily represent an effective antitumor response; rather, they may indicate a complex state involving amplified inflammation, immune remodeling, or even immune exhaustion [14, 34, 35]. Elevated neutrophils may release neutrophil extracellular traps and promote the colonization of circulating tumor cells; increased monocytes may induce polarization of tumor‐associated macrophages toward the M2 phenotype, thereby suppressing antitumor immunity [36, 37, 38, 39]. Conversely, lymphocytes are essential components of antitumor immune responses, and lymphopenia often indicates impaired cellular immunity, potentially weakening the clearance of minimal residual disease and circulating tumor cells [40, 41]. Hypoalbuminemia and malnutrition may also impair cellular immune function and limit the host's ability to eliminate micrometastatic disease [42, 43]. Importantly, however, postoperative day‐7 indicators should not be interpreted in the same manner as their preoperative counterparts, because they represent composite measures influenced not only by tumor biology but also by surgical trauma, the extent and radicality of resection, postoperative complications, infection, perioperative treatment, and individual recovery. Consequently, the prognostic direction of high‐ versus low‐level groups may differ before and after surgery. For example, patients undergoing more extensive or radical surgery may experience a stronger postoperative inflammatory response and a higher NLR while simultaneously achieving better long‐term tumor control. These perioperative factors may therefore affect postoperative biomarker levels and oncological outcomes in different, or even opposite, directions and contribute to residual confounding. Taken together, these considerations highlight the importance of dynamically evaluating inflammatory–immune–nutritional indicators across perioperative time points rather than relying on a single preoperative measurement.

Among the indicators evaluated in this study, NLR, MLR, and SIRI mainly reflect systemic inflammation and changes in immune cell composition, whereas PNI, ALI, AAPR, AGR, and ALBI more strongly reflect nutritional reserve, liver function, and metabolic burden [18, 19, 20, 21, 22, 23, 24]. Preoperative indicators primarily represent the baseline condition before treatment, including chronic inflammation, nutritional reserve, and liver function. Postoperative indicators further integrate surgical trauma, perioperative stress, risk of complications, and recovery capacity. In addition, postoperative Day 7 is usually close to clinical discharge and the time when subsequent adjuvant treatment plans are formulated. Laboratory indicators at this time point may more accurately reflect the patient's true status during early postoperative recovery and before adjuvant therapy. Therefore, inflammatory–immune–nutritional indicators measured at postoperative Day 7 are not merely laboratory values; they may also reflect whether a patient can recover smoothly, tolerate subsequent treatment, and avoid early recurrence.

This study showed that the eight perioperative inflammatory–immune–nutritional indicators changed significantly from the preoperative period to postoperative Day 7. These dynamic changes visually reflect the pathophysiological processes of systemic inflammatory activation, immune suppression, and short‐term nutritional decline induced by surgical stress, which may be closely related to micrometastasis and the survival of circulating tumor cells. Patients who are classified as high risk based on preoperative inflammatory–immune–nutritional indicators may require more active perioperative nutritional support, inflammation control, liver function optimization, and multidisciplinary assessment. Similarly, patients who continue to show persistent inflammation or marked deterioration in nutritional or immune status on postoperative Day 7 should be monitored carefully for postoperative complications, delayed adjuvant therapy, or early recurrence.

In clinical practice, postoperative Day 7 represents a critical window of perioperative stress, inflammatory response, and physical recovery. Although this time point is influenced by surgical trauma and perioperative management, precisely because of this, it may sensitively capture the biological response to surgical insult and the interaction between tumor and host [44]. It therefore compensates for the inability of a single preoperative indicator to reflect postoperative biological status.

After preoperative indicators, postoperative indicators, and combined preoperative plus postoperative dynamic indicators were incorporated into models based on clinical and pathological variables, most dynamic indicators improved discrimination and goodness of fit to varying degrees. This suggests that perioperative dynamic information may more closely approximate the true prognostic state of the patient than information from a single time point, consistent with findings in other cancer types [18, 19, 20, 21, 22, 23, 24]. These results indicate that survival outcomes in advanced GBC are not determined solely by tumor stage and treatment modality; the patient's inflammatory burden, immune status, and nutritional reserve also influence disease progression and treatment outcomes.

Patients with advanced GBC often have appetite loss, biliary obstruction, chronic inflammation, and a high prevalence of malnutrition before surgery [14]. Surgical trauma further increases metabolic consumption, leading to delayed initiation of adjuvant therapy, insufficient dose intensity, increased treatment interruption, higher rates of postoperative complications and infection, longer hospitalization, and indirectly reduced benefits from multimodality therapy [45, 46, 47]. These findings suggest that perioperative management should not focus solely on tumor resection. Early nutritional support, inflammation control, and immune modulation may improve indicator levels, increase treatment completion, and ultimately translate into survival benefit. This provides a clear intervention direction for perioperative comprehensive management. Moreover, because this evaluation framework is based on routine blood tests, it does not require additional costs or increase patient burden, making it suitable for use in primary hospitals and large‐scale clinical practice. It may also compensate for the lack of simple and dynamic prognostic tools for advanced GBC. Future incorporation of longer follow‐up data may further improve the completeness and accuracy of the model.

Several limitations should be acknowledged. First, this was a single‐center retrospective study with a relatively small sample size and only 47 cancer‐specific deaths, which may have introduced selection bias, limited the precision and generalizability of the estimates, and increased the risk of overfitting. In particular, the baseline and fully expanded models contained eight and ten predictor terms, respectively, resulting in a relatively low events‐to‐parameter ratio. Although the baseline covariates were prespecified because of their established clinicopathological relevance and were retained to avoid underadjustment and overestimation of the incremental contribution of the inflammatory–immune–nutritional indicators, their inclusion does not eliminate the instability associated with the limited number of events. Second, the cutoff values were derived and evaluated within the same cohort and may therefore be subject to optimism and cohort‐specific effects. Although bootstrap resampling was used to assess their internal stability, these thresholds should be regarded as exploratory rather than established clinical cutoffs because no independent external validation cohort was available. Third, perioperative inflammatory–immune–nutritional indicators may be influenced by surgical trauma, infection, biliary obstruction, postoperative complications, fluid administration, liver function changes, nutritional support, and individual recovery. In addition, chemotherapy and radiotherapy were recorded only as binary variables, and residual confounding related to treatment regimen, intensity, timing, and completion may have affected both CSS and the observed biomarker–prognosis associations. Finally, this study primarily evaluated statistical associations and incremental model performance and did not establish whether these indicators improve clinical decision‐making or patient outcomes, nor did it elucidate their underlying biological mechanisms. Therefore, the present findings should be considered exploratory and hypothesis‐generating, and further validation in larger prospective multicenter cohorts with standardized measurements, sufficient numbers of events, appropriate shrinkage or penalization, and independent external validation, together with mechanistic studies, is warranted.

5. Conclusion

Perioperative inflammatory–immune–nutritional indicators show significant dynamic changes in patients with advanced GBC. Several indicators were associated with CSS, and the addition of individual preoperative or postoperative indicators was associated with modest improvements in the performance of clinicopathological models in the present cohort. These findings suggest that dynamic inflammatory–immune–nutritional indicators may serve as important supplementary tools for risk stratification and individualized prognostic assessment in patients with advanced GBC.

Author Contributions

Rongxuan Li: writing – original draft, conceptualization, methodology, software, formal analysis, investigation, writing – review and editing, data curation, visualization, validation. Bingchen Wang: conceptualization, software, formal analysis, investigation, writing – review and editing. Xiao Chen: conceptualization, methodology, investigation, writing – review and editing, formal analysis. Dayong Cao: supervision, resources, project administration. Jianguo Zhou: supervision, resources, project administration.

Funding

The authors have nothing to report.

Ethics Statement

The studies involving humans were approved by the Ethics Committee of the National Cancer Center/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College (approval no. 23/285–4027). The studies were conducted in accordance with local legislation and institutional requirements. Written informed consent for participation was not required from participants or the participants' legal guardians/next of kin in accordance with national legislation and institutional requirements.

Consent

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Pairwise Spearman correlations among preoperative and postoperative inflammatory–immune–nutritional indicators.

Table S1: Optimal cutoffs and survival associations of perioperative inflammatory–immune–nutritional indicators.

Table S2: Bootstrap internal validation of outcome‐derived cutoff values for perioperative inflammatory–immune–nutritional indicators.

Figure S2: Calibration and decision curve analyses of the clinicopathological baseline and indicator‐extended prognostic models in advanced GBC.

Figure S3: Preoperative and postoperative comparisons of the eight inflammatory–immune–nutritional indicators in the Stage III subgroup of the institutional cohort.

Figure S4: CSS curves after dichotomization of the eight preoperative inflammatory–immune–nutritional indicators in the Stage III subgroup of the institutional cohort.

Figure S5: CSS curves after dichotomization of the eight postoperative inflammatory–immune–nutritional indicators in the Stage III subgroup of the institutional cohort.

Table S3: Model comparison after combining the eight inflammatory–immune–nutritional indicators with clinical and pathological variables in the Stage III subgroup of the institutional cohort.

Figure S6: Preoperative and postoperative comparisons of the eight inflammatory–immune–nutritional indicators in the Stage IV subgroup of the institutional cohort.

Figure S7: CSS curves after dichotomization of the eight preoperative inflammatory–immune–nutritional indicators in the Stage IV subgroup of the institutional cohort.

Figure S8: CSS curves after dichotomization of the eight postoperative inflammatory–immune–nutritional indicators in the Stage IV subgroup of the institutional cohort.

Table S4: Model comparison after combining the eight inflammatory–immune–nutritional indicators with clinical and pathological variables in the Stage IV subgroup of the institutional cohort.

Figure S9: Preoperative and postoperative comparisons of the eight inflammatory–immune–nutritional indicators in the complete‐case cohort.

Figure S10: CSS curves after dichotomization of the eight preoperative inflammatory–immune–nutritional indicators in the complete‐case cohort.

Figure S11: CSS curves after dichotomization of the eight postoperative inflammatory–immune–nutritional indicators in the complete‐case cohort.

Table S5: Model comparison after combining the eight inflammatory–immune–nutritional indicators with clinical and pathological variables in the complete‐case institutional cohort.

Data Availability Statement

All raw data and code are available upon request.