Comparison of Baseline Systemic Inflammatory Indicators in Preoperative Treatment of Gastric Carcinoma

1. Introduction

Gastric cancer is the fifth most common malignancy globally and the fourth leading cause of cancer-related mortality, with approximately 968,000 new cases and 660,000 deaths annually [1]. Although localized disease is associated with a relatively favorable prognosis, patients with regional lymph node involvement have a five-year survival rate of only 37%, highlighting the critical need for effective multimodal treatment strategies [2]. The landmark MAGIC trial transformed perioperative management by demonstrating that the addition of epirubicin, cisplatin, and fluorouracil to surgery significantly improves survival in resectable gastric and gastroesophageal junction (GEJ) cancer [3]. Subsequently, the FLOT4-AIO trial established FLOT (5-fluorouracil, leucovorin, oxaliplatin, and docetaxel) as the standard of care, surpassing ECF/ECX-based regimens and extending median overall survival beyond four years in locally advanced cases [4]. More recently, the MATTERHORN trial demonstrated that the addition of the PD-L1 inhibitor durvalumab to perioperative FLOT enhances pathological complete response and event-free survival [5]. However, the substantial cost of immunotherapy remains a significant barrier for healthcare systems and patients, underscoring the necessity of careful patient selection [6].

Systemic inflammatory and nutritional indices—calculated from routine blood counts and serum biochemistry—have therefore emerged as practical, low-cost surrogates for the host immune microenvironment and systemic inflammatory status. Their prognostic utility is grounded in the well-established role of cancer-related inflammation in facilitating tumor immune evasion, promoting angiogenesis, and enabling metastatic dissemination [7]. Large-scale meta-analyses have since established the prognostic value of indices such as the neutrophil-to-lymphocyte ratio (NLR) and the Prognostic Nutritional Index (PNI)—the latter originally developed to predict postoperative complications in malnourished surgical patients [8,9]—across multiple gastrointestinal malignancies [10].

Although the prognostic significance of individual inflammatory markers is well established, their comparative performance in the perioperative chemotherapy context for gastric cancer remains insufficiently characterized [11]. Notably, few studies have concurrently assessed a comprehensive panel of nine composite indices—including recently developed scores such as SII, SIRI, HALP, ALBI, and PALBI—in relation to pathological tumor regression and long-term survival in the neoadjuvant setting against pathological tumor regression and long-term survival in the neoadjuvant setting [12]. To address this knowledge gap, the present study systematically evaluated these indices in a real-world gastric cancer cohort to identify the blood-based biomarker most reliably associated with Mandard Tumor Regression Grade (TRG) and clinical outcomes.

2. Materials and Methods

2.1. Study Design and Population

This retrospective study collected patient data from two oncology centers in Izmir, Turkey—the Izmir City Hospital and the Dokuz Eylül University Faculty of Medicine—between 2020 and 2025. Patients were eligible if they met all of the following criteria: age ≥ 18 years; histopathologically confirmed gastric or gastro-oesophageal junction (GEJ) adenocarcinoma; locally advanced, resectable disease (cT2–T4 and/or cN+, M0) confirmed by preoperative computed tomography (CT) staging; case reviewed and approved for perioperative systemic therapy at the multidisciplinary tumor board; receipt of at least one cycle of neoadjuvant chemotherapy followed by curative-intent (R0) gastrectomy; availability of baseline laboratory values (complete blood count and serum biochemistry) obtained within seven days before initiating systemic therapy; and a complete postoperative histopathological report including Mandard TRG, ypT stage, and ypN stage. Patients were excluded if they had synchronous or metachronous second primary malignancies, distant metastasis (M1) at the time of diagnosis or prior to surgery, non-curative surgical intent, or insufficient baseline laboratory data to calculate at least one inflammatory index. As this was an observational retrospective study, no control arm was included; the study was designed to reflect real-world clinical outcomes in a middle-income healthcare setting.

2.2. Data Sources and Variables

Clinical data were extracted from electronic medical records and institutional databases. Collected variables included age, sex, tumor location, histologic subtype, tumor grade, biomarker status (EBV, MMR, HER2, p53), details of the perioperative treatment regimen (neoadjuvant and adjuvant regimens, number of cycles, dose modifications), surgical records (extent of lymphadenectomy, number of nodes harvested), and postoperative histopathologic features, including Mandard TRG, ypT stage, and ypN stage.

Baseline laboratory values were defined as tests obtained within seven days before the initiation of systemic therapy for a pathologically confirmed diagnosis of gastric cancer. The following nine inflammatory and nutritional indices were calculated from routine complete blood count and serum biochemistry parameters:

Neutrophil-to-lymphocyte ratio (NLR) = neutrophil count / lymphocyte count; monocyte-to-lymphocyte ratio (MLR) = monocyte count / lymphocyte count; neutrophil-to-monocyte ratio (NMR) = neutrophil count / monocyte count; systemic immune-inflammation index (SII) = (platelet count × neutrophil count) / lymphocyte count; systemic inflammatory response index (SIRI) = (monocyte count × neutrophil count) / lymphocyte count; Prognostic Nutritional Index (PNI) = (10 × serum albumin [g/dL]) + (0.005 × total lymphocyte count [/mm3]); hemoglobin–albumin–lymphocyte–platelet (HALP) score = hemoglobin [g/L] × albumin [g/L] × lymphocyte count [/L] / platelet count [/L]; albumin–bilirubin index (ALBI) = (0.66 × log10[total bilirubin (μmol/L)]) + (−0.085 × albumin [g/L]); and platelet–albumin–bilirubin index (PALBI) = (2.02 × log10 bilirubin) − (0.37 × [log10 bilirubin]2) − (0.04 × albumin) − (3.48 × log10 platelets) + (1.01 × log10 platelets). When laboratory units were reported in non-SI format, standard conversions were applied: bilirubin (mg/dL) × 17.1 = μmol/L; albumin (g/dL) × 10 = g/L.

2.3. Outcomes

The primary endpoint was the association between pre-treatment PNI values and major pathological response, defined as Mandard TRG 1–2. Disease-free survival (DFS) was the time from first cycle of systemic therapy to the first documented locoregional recurrence, distant metastasis, or death from any cause, whichever occurred first. Overall survival (OS) was the time from diagnosis to death from any cause.

2.4. Statistical Analysis

Statistical analyses were conducted using IBM SPSS Statistics, version 29. 0. Continuous variables were assessed for normality with the Kolmogorov–Smirnov test and are reported as median (range) or mean ± standard deviation, as appropriate. Categorical variables are reported as frequency and percentage and compared by chi-square or Fisher’s exact test (the latter applied when expected cell counts fell below 5). Pre- to post-neoadjuvant changes in inflammatory markers were compared using the Wilcoxon signed-rank test, given non-normal distributions confirmed by the Kolmogorov–Smirnov test. Associations between baseline indices and TRG across the full five-level spectrum were examined with the Kruskal–Wallis test; post hoc pairwise comparisons used the Mann–Whitney U test with Bonferroni correction. For the dichotomized primary outcome (TRG 1–2 vs. TRG 3–5), the Mann–Whitney U test was applied. Receiver operating characteristic (ROC) curve analysis was performed to assess the discriminatory ability of each index, with optimal cut-off values derived using the Youden index (J = sensitivity + specificity − 1). For biomarkers where the Youden-optimal threshold predicted the outcome in the inverse direction (i.e., lower values associated with the event), the effective AUC was calculated as 1 − AUC. To address optimism bias inherent in deriving cut-off values from the same dataset used for performance estimation, internal validation of PNI-based ROC analyses was performed via non-parametric bootstrap resampling with 1,000 replicates (random seed fixed for reproducibility). For each bootstrap sample, the AUC and the Youden-optimal cut-off were re-derived; bias-corrected and accelerated (BCa) 95% confidence intervals were computed for the AUC, and the empirical distribution of bootstrap-derived cut-offs (median and 2.5th–97.5th percentile range) was used to assess threshold stability. Optimism in the Youden index was estimated as the mean difference between in-sample (bootstrap) and out-of-sample (original cohort) values of (sensitivity + specificity), and the optimism-corrected Youden index was calculated as the apparent value minus the mean optimism. Median follow-up was estimated using the reverse Kaplan–Meier method. Univariable and multivariable Cox proportional hazards regression were performed for DFS and OS. Multivariable logistic regression was performed for pathological response; this model was fitted to patients with complete covariate data (N = 119). The 11 patients excluded had missing values for one or more covariates and did not differ significantly in baseline characteristics from the complete-case population. Variables entered into multivariable models were those with clinical relevance or a p-value < 0.10 in univariable analyses. The proportional hazards assumption was verified using log-minus-log plots. Effect sizes were expressed as odds ratios (ORs) or hazard ratios (HRs) with 95% confidence intervals (CIs). A two-sided p-value < 0.05 was considered statistically significant. Patients who had not experienced a defining event were censored at the date of their last clinical follow-up. Survival probabilities were estimated using the Kaplan–Meier method, and differences between groups were assessed with the log-rank test.

2.5. Ethics Statement

This study was conducted in accordance with the ethical standards of the Declaration of Helsinki. The research protocol was reviewed and approved by the institutional review boards of all participating centers, with primary ethical approval granted by the Izmir City Hospital Ethics Committee (Decision number: 2025/159; Date: 16 April 2025). Written informed consent for both the prescribed treatment and the retrospective use of anonymized clinical data was obtained from all patients before therapy began.

3. Results

3.1. Baseline Characteristics

One hundred and thirty patients with gastric adenocarcinoma met the eligibility criteria. Their median age was 63 years (range 28–83); 90 patients (69.2%) were male. The primary tumor was located in the proximal stomach (GEJ and cardia) in 45 patients (34.6%), the corpus in 57 (43.8%), and the distal stomach (antrum) in 28 (21.5%). By histological subtype, mixed or not otherwise specified (NOS) was the most common (n = 53, 40.8%), followed by diffuse (n = 36, 27.7%), intestinal (n = 33, 25.4%), and mucinous carcinomas (n = 8, 6.2%). Most patients received FLOT as the neoadjuvant backbone (n = 124, 95.4%), with six patients (4.6%) receiving FOLFOX. Full baseline characteristics, stratified by TRG group, are presented in Table 1. No significant differences in clinicopathological or treatment characteristics were found between TRG groups at baseline (p > 0.05).

3.2. Pathologic Response

TRG was strongly associated with post-treatment pathological stage: both ypT (p < 0.001) and ypN distributions (p = 0.004) differed significantly between the TRG 1–2 and TRG 3–5 groups. Pathological complete response (ypT0N0) occurred exclusively in the TRG 1–2 group (18.8%; p < 0.001). Creatinine was the only additional laboratory parameter significantly higher in the TRG 1–2 group (p = 0.024), whereas the albumin–globulin ratio showed a borderline difference (p = 0.052).

Among all continuous inflammatory and nutritional indices, only pre-treatment PNI differed significantly between response groups, being higher in TRG 1–2 patients (median 53.5 vs. 50.1; p = 0.047). In ROC analyses, albumin (AUC 0.633, SE 0.062, 95% CI 0.512–0.755; p = 0.034) and PNI (AUC 0.645, SE 0.064, 95% CI 0.519–0.771; p = 0.021) showed statistically significant discriminatory performance. Using the Youden index, the optimal PNI cut-off was ≥ 53.125 (sensitivity 57.1%, specificity 73.6%). Internal validation by 1,000-resample bootstrap (BCa method) yielded a mean AUC of 0.617 (SD 0.066) with a 95% CI of 0.490–0.743; the median bootstrap-derived cut-off was 53.8 (95% range 50.0–57.0), confirming that the apparent threshold of 53.125 lies within a stable region of the operating characteristic. The optimism in the Youden index was 0.057, yielding an optimism-corrected Youden value of 1.227 (compared with the apparent value of 1.283), and mean operating characteristics of approximately 49% sensitivity and 77% specificity when the bootstrap-derived cut-off was applied to the original cohort. All other indices—NLR, MLR, NMR, SII, SIRI, HALP, ALBI, and PALBI—had AUCs not significantly different from 0.50 (all p > 0.05).

In univariable logistic regression, PNI ≥ 53.125 was significantly associated with TRG 1–2 (OR 3.429, 95% CI 1.461–8.047; p = 0.005), and PNI as a continuous variable was also significant (OR 1.089 per unit, 95% CI 1.007–1.178; p = 0.034). Age, gender, tumor location, histologic subtype, grade, MMR status, FLOT exposure, dose reduction, and all other inflammatory indices were not significant predictors. CA19-9 ≥ 37 showed a borderline inverse association with TRG status (OR 0.302; p = 0.066). In the multivariable logistic regression model (n = 119), PNI ≥ 53.125 remained the sole independent predictor of major pathological response (adjusted OR, 4.200; 95% CI, 1.631–10.820; p = 0.003) (Table 2). CA19-9 status, number of FLOT cycles, and signet-ring cell component were not independently significant.

3.3. Disease-Free Survival

During a median follow-up of 22.2 months (reverse Kaplan–Meier), 37 events were observed. Median DFS was 30.6 months (95% CI 23.0–38.3). Recurrence was significantly less frequent in the TRG 1–2 group (12.5%) than in the TRG 3–5 group (33.7%) (p = 0.024). In Kaplan–Meier analysis, TRG 1–2 patients had significantly longer DFS than TRG 3–5 patients (median DFS not reached vs. 27.6 months [95% CI 14.4–40.8]; log-rank p = 0.033).

In the ROC analysis for recurrence, PNI was the only biomarker with a statistically significant AUC (AUC 0.378, SE 0.060, 95% CI 0.261–0.496; p = 0.043). Because an AUC below 0.50 indicates inverse discrimination, lower PNI was interpreted as associated with recurrence risk, yielding an effective discriminatory AUC of 1 − 0.378 = 0.622. The Youden-optimal threshold for predicting recurrence was PNI ≤ 48.475 (sensitivity 59.4%, specificity 68.7%). Internal validation by 1,000-resample bootstrap (BCa method) yielded a mean effective AUC of 0.611 (SD 0.058) with a 95% CI of 0.499–0.730; the median bootstrap-derived cut-off was PNI ≤ 49.2, with a 95% range of 41.9–54.1, indicating moderate threshold variability around the apparent value of 48.475. The optimism in the Youden index was 0.075, with an optimism-corrected Youden value of 1.159 (apparent value 1.233) and mean operating characteristics of approximately 58% sensitivity and 62% specificity when the bootstrap-derived cut-off was applied to the original cohort. All other indices showed no significant discriminatory ability for recurrence (p > 0.05).

In univariable Cox regression, TRG 1–2 was associated with improved DFS (HR 0.338, 95% CI 0.119–0.960; p = 0.042). Post-treatment nodal stage was strongly prognostic (ypN overall p < 0.001), with ypN2 (HR 3.116; p = 0.037) and ypN3 (HR 5.874; p < 0.001) associated with significantly higher recurrence hazard than ypN0. Among biomarkers, PNI was significantly associated with improved DFS as a continuous variable (HR 0.937 per unit, 95% CI 0.889–0.987; p = 0.015) and as a dichotomous variable (high vs. low PNI: HR 0.488, 95% CI 0.250–0.952; p = 0.035). Completion of perioperative therapy was also strongly protective in univariable analysis (HR 0.272; p < 0.001).

In the multivariable Cox model for DFS (n = 126; 36 events), only completion of perioperative treatment remained an independent predictor of improved DFS (HR 0.386, 95% CI 0.179–0.832; p = 0.0144). TRG 1–2 showed a non-significant trend toward better DFS (HR 0.336, 95% CI 0.101–1.119; p = 0.064), whereas PNI was not independently significant (HR 0.962 per unit, 95% CI 0.912–1.015; p = 0.147) (Table 3).

3.4. Overall Survival

A total of 34 deaths (26.2%) occurred during a median follow-up of 22.1 months. The median OS was 38.8 months (95% CI 20.6–57.0). In Kaplan–Meier analysis, OS did not differ significantly between TRG subgroups: the median OS was 33.2 months (95% CI 32.1–34.2) for TRG 1–2 versus 38.8 months (95% CI 16.3–61.3) for TRG 3–5 (log-rank p = 0.316). In univariable Cox regression, post-treatment nodal stage (ypN, overall p = 0.007) and completion of perioperative treatment (HR 0.424, 95% CI 0.196–0.918; p = 0.030) were associated with OS, whereas TRG (HR 0.637; p = 0.320) and inflammatory and nutritional indices—including PNI, NLR, NMR, LMR, SII, SIRI, HALP, ALBI, and PALBI—were not. In the multivariable Cox model, ypN3 (vs. ypN0: HR 3.689, 95% CI 1.563–8.708; p = 0.003) and completion of perioperative therapy (HR 0.428, 95% CI 0.195–0.943; p = 0.035) remained independently associated with OS (Table 4).

4. Discussion

4.1. Benchmarking Against Real-World and Trial Data

This study provides real-world evidence supporting the prognostic utility of pre-treatment inflammatory markers, particularly PNI, in gastric cancer patients undergoing perioperative chemotherapy and curative-intent surgery [13,14,15]. The cohort’s median OS of 38.8 months and DFS of 30.6 months are consistent with contemporary real-world FLOT series, which typically report median OS between 35 and 42 months. The modest difference compared to the FLOT4-AIO trial (median OS 50 months) likely reflects older patient demographics, a higher comorbidity burden, and the unselected nature of routine clinical practice [16]. The importance of real-world attrition is further highlighted by the finding that completion of the full perioperative regimen was among the strongest independent predictors of both DFS and OS, in agreement with international observational data [17].

Pathological response rates confirmed the efficacy of the administered neoadjuvant regimens. The pCR rate of 18.8% in the TRG 1–2 subgroup is consistent with the 16% pCR rate reported in FLOT4 [4]. Achieving TRG 1–2 was significantly associated with lower recurrence (12.5% vs. 33.7%; p = 0.024) and longer DFS in Kaplan–Meier analysis (log-rank p = 0.033), supporting TRG as a clinically meaningful surrogate endpoint of treatment efficacy [18]. Our cohort’s survival and pathological outcomes align with global standards, providing a solid and representative baseline for our biomarker analysis.

Despite superior DFS, TRG 1–2 patients exhibited a numerically shorter median OS (33.2 vs. 38.8 months; p = 0.316) compared to the TRG 3–5 group. This apparent paradox likely results from data immaturity at 22.1 months of follow-up, the small size of the TRG 1–2 subgroup (n = 32), and the unusually narrow OS confidence interval in this group (95% CI 32.1–34.2), which is indicative of early censoring rather than true event accrual. With extended follow-up, the OS benefit associated with major pathological response—already evident in the DFS curves—is expected to become apparent, consistent with findings from other perioperative series with more mature data

4.2. Comparative Analysis of Systemic Inflammatory Indices

The central finding is that PNI was the only pre-treatment index independently predictive of major pathological response (TRG 1–2) in multivariable analysis and the only biomarker with significant discriminatory performance for both pathological response (AUC 0.645; p = 0.021) and recurrence (effective AUC 0.622; p = 0.043) in ROC analysis. In univariable Cox modeling for DFS, PNI was significantly protective (HR 0.937 per unit; p = 0.015; high vs. low: HR 0.488; p = 0.035). However, this association attenuated in the multivariable DFS model (p = 0.160), where only perioperative treatment completion retained independent significance. That attenuation is clinically fair: patients with higher baseline PNI—reflecting better nutritional reserve and immune competence—are simply better placed to complete the full treatment course, and perioperative completion likely mediates a portion of the PNI effect on DFS. Adjusting for perioperative completion thus absorbs much of the PNI effect on DFS, rendering the direct PNI–DFS pathway non-significant in the adjusted model.

The superiority of PNI over purely inflammatory indices such as NLR, SII, and SIRI is mechanistically intuitive. PNI uniquely integrates two determinants of anti-tumor immunity: nutritional reserve, indexed by serum albumin—which sustains lymphocyte proliferation, cytokine production, and the hepatic acute-phase response—and adaptive immune competence, reflected by the peripheral lymphocyte count [20,21,22,23,24]. Neutrophilia and monocytosis—the primary components of NLR, SII, and SIRI—reflect innate immune activation and are susceptible to confounding by transient, non-oncological states such as intercurrent infection or corticosteroid use [25,26]. Albumin, by contrast, provides a more stable reflection of systemic nutritional status, resists acute-phase fluctuations, and likely confers greater discriminatory stability across the neoadjuvant treatment period [27].

Neither the ALBI nor the PALBI grade showed significant associations with oncological endpoints, consistent with the premise that these indices—originally developed as hepatic reserve surrogates in hepatocellular carcinoma [28]—have limited discriminatory power in non-cirrhotic gastric cancer populations, where bilirubin variability is narrow. The HALP score similarly failed to reach significance, possibly because the platelet component introduces counter-directional noise in patients without thrombocytosis [29]. Taken together, these findings confirm that composite inflammatory scores do not perform uniformly across tumor types and treatment settings—and that biological context, not convenience, should guide index selection.

4.3. Limitations

This study has several limitations. The retrospective design introduces inherent risks of selection bias and residual confounding not fully addressable by multivariable adjustment. The sample size of 130 patients limits statistical power, particularly for subgroup analyses and modest OS associations, with only 34 events recorded at current follow-up. The use of two neoadjuvant regimens—predominantly FLOT (95.4%) and a small proportion of FOLFOX (4.6%)—introduces therapeutic heterogeneity, though their distribution was balanced across TRG groups. A median follow-up of approximately 22 months is insufficient for definitive OS conclusions, as reflected in the paradoxically higher numerical OS in the TRG 3–5 group—an artifact of immature curves and differential censoring. The study was conducted in a middle-income healthcare setting, and differences in perioperative support and toxicity management may limit generalizability to higher-resourced systems. Tumor regression grading was performed locally at each institution without central pathological review, introducing potential interobserver variability in TRG classification. Inflammatory and nutritional indices were assessed at baseline only; longitudinal post-neoadjuvant profiling—which could better capture treatment-induced immunomodulation—was not systematically performed. Finally, the PNI cut-off values were derived from the same cohort in which they were evaluated. To mitigate the resulting optimism bias, internal validation was performed using a 1,000-resample bootstrap with bias-corrected and accelerated (BCa) confidence intervals; the apparent thresholds remained within stable bootstrap-derived ranges, and the optimism in the Youden index was modest (0.057 for the response endpoint and 0.075 for the recurrence endpoint). Nonetheless, internal validation cannot fully substitute for prospective external validation in geographically and clinically distinct cohorts, which remains required before clinical implementation.

4.4. Clinical Implications

That PNI emerged as the sole pre-treatment biomarker independently associated with major pathological response in the multivariable analysis, together with its significant univariable association with DFS and the ROC-derived cut-offs, positions it as a practical and accessible stratification tool for perioperative gastric cancer management. Crucially, PNI requires only two universally available and inexpensive measurements—serum albumin and peripheral lymphocyte count—making it feasible to implement in virtually any clinical setting worldwide. A pretreatment PNI of ≥ 53.125 may flag patients who are nutritionally and immunologically primed, and therefore most likely to achieve meaningful pathological downstaging after neoadjuvant chemotherapy. Conversely, a pretreatment PNI ≤ 48.475 identifies a high-risk subgroup with substantially elevated recurrence risk, pointing toward closer postoperative surveillance and consideration of treatment intensification or alternative adjuvant approaches.

These findings are directly relevant to the current era of perioperative immunotherapy. The MATTERHORN trial has established the benefit of adding durvalumab to FLOT [5], but the cost-effectiveness of universal immunotherapy adoption remains uncertain [6]. Our findings suggest that pretreatment PNI could help inform patient selection for escalated regimens: patients with low PNI may lack the immunological substrate for effective checkpoint inhibitor responses and may simultaneously bear a disproportionate toxicity burden given their compromised nutritional reserves. Prospective evaluation of PNI as a predictive—rather than solely prognostic—biomarker for immunotherapy benefit is warranted.

The finding that perioperative treatment completion was the strongest independent predictor of both DFS and OS reinforces the importance of proactive nutritional support, toxicity management, and prehabilitation throughout the treatment course. Because baseline PNI captures both nutritional reserve and immune competence, it may also predict a patient’s ability to tolerate the full treatment course. Formal prospective evaluation of nutritional interventions targeting PNI—before and during neoadjuvant chemotherapy—is warranted, with PNI serving as both a stratification variable and an intermediate endpoint.