Residual Lymph Nodes After Immunotherapy Predict Gastric Cancer Survival, Study Finds

For patients with locally advanced gastric cancer, the journey rarely ends in the operating theater. Most now arrive at surgery carrying the imprint of months of drug therapy—chemotherapy paired with immune checkpoint inhibitors—administered before any blade touches tissue. This strategy, known as neoadjuvant immunochemotherapy, has transformed outcomes for many, shrinking tumors and awakening the immune system against cancer cells. Yet a stubborn question has shadowed the field: once the tumor is removed, how precisely should doctors judge each patient’s remaining risk? A new retrospective cohort study published in BMC Cancer offers a technically elegant answer, arguing that two pieces of pathological information—the degree of tumor regression and the exact number of cancer-positive lymph nodes left behind—work together to stratify survival far better than conventional staging alone.

The research, led by Meng Yu, Yapeng Qi, Gaofeng Huang, Jindai Zhang, Chunguang Li, and Zhimeng Li, drew on 286 patients with locally advanced gastric or Siewert type III gastroesophageal junction adenocarcinoma. All had received neoadjuvant immunochemotherapy followed by curative-intent gastrectomy between 2018 and 2024. The investigators split the population geographically: 210 patients treated at Henan Cancer Hospital formed the development cohort, where statistical models were built, while 76 patients from Taikang People’s Hospital served as an independent geographic validation cohort, with model coefficients locked before testing. This two-cohort architecture matters because prediction models in oncology have a notorious habit of flattering their own data; testing on patients from a different hospital provides a sterner, more honest examination.

The study’s central innovation lies in how it quantifies residual disease in the lymph nodes. Traditional pathology reports assign a category—ypN0, ypN1, ypN2, and so on—based on how many nodes contain metastatic cancer. But categories collapse information. A patient with one positive node and a patient with seven are lumped into neighboring bins, and the clinical gulf between them can be enormous. The researchers instead modeled residual nodal burden as a continuous variable, transformed as the base-two logarithm of the positive-node count plus one. The logarithm is not statistical decoration. It means each one-unit increase in the predictor corresponds to a doubling of the positive-node count plus one, capturing the steeply diminishing survival curve that clinicians observe as nodal disease accumulates, while keeping extreme counts from dominating the model.

Over a median follow-up of 36.8 months, the cohort generated 92 disease-free survival events and 59 deaths—enough events to power multivariable Cox regression with reasonable stability. The results were strikingly consistent. In the development cohort, each doubling of the positive-node count plus one carried an adjusted hazard ratio of 1.58 for disease-free survival, with a 95 percent confidence interval of 1.34 to 1.86. In the validation cohort, the hazard ratio was 1.49 (1.14 to 1.95). For overall survival, the associations were similarly robust: 1.66 (1.35 to 2.04) in the development cohort and 1.57 (1.12 to 2.20) in validation. In plain terms, every doubling of residual nodal disease cut the expected time without recurrence by a substantial margin, independent of tumor depth, nodal category, and other covariates in the models.

The second axis of the framework is primary-tumor pathological response. Pathologists examined the resected primary tumor and classified regression as either major pathological response—indicating that treatment had eradicated nearly all viable cancer cells in the specimen—or non-major response. This distinction has become increasingly meaningful in the immunochemotherapy era, because immune checkpoint inhibitors can produce deep, durable tumor regressions that chemotherapy alone rarely achieves. A major response signals a biologically hot tumor, one whose microenvironment responded vigorously to immunologic stimulation. The study treated this binary classification and the continuous nodal burden as complementary rather than competing signals: one describes what the drugs did to the tumor, the other describes what disease the surgeon actually removed.

When the two variables were combined, the survival gradient became dramatic. Three-year disease-free survival ranged from 90.4 percent among patients who achieved a major pathological response and had no residual nodal disease—the best of both worlds—down to 51.6 percent among patients with neither a major response nor nodal clearance. That nearly forty-percentage-point spread, within a single postoperative population, illustrates why the authors argue that conventional pathological staging, which relies on tumor and nodal categories, leaves clinically actionable information on the table. Two patients with identical ypT and ypN categories can sit at opposite ends of that survival spectrum depending on their tumor regression and their exact residual node count.

To test whether the new variables genuinely improved prediction, the investigators used nested model comparisons built on pathological T and N categories. Adding the continuous nodal burden increased the concordance index—the C-index, which measures how well a model ranks patients by risk—by 0.034 (95 percent confidence interval, 0.006 to 0.063) for disease-free survival beyond pathological T and N categories alone. Adding it on top of the categories plus primary-tumor response still yielded a gain of 0.030 (0.004 to 0.057). These increments are modest in absolute terms but meaningful for a validated model, and the confidence intervals excluding zero indicate the improvement is unlikely to be statistical noise. The full model achieved a C-index of 0.721 in the geographic validation cohort, a level of discrimination that compares favorably with many established oncology prognostic tools.

Methodological rigor underpins these claims. Internal validation used 1,000 bootstrap resamples, a technique that repeatedly refits the model on resampled data to estimate how much its performance is inflated by overfitting. The decision to lock coefficients before geographic validation prevents the subtle data leakage that can occur when model parameters are tuned on the same patients used to evaluate them. The prespecified log transformation of the nodal count, illustrated in the study’s supplementary materials, ensures the predictor’s interpretation remains transparent: a one-unit change is always a doubling. Ethics approval was obtained from both participating hospitals, written informed consent was collected from all patients at the initiation of treatment, and the authors declared no competing interests and no dedicated funding.

The clinical implications are considerable, even if the authors are careful about timing. If residual positive-node count and tumor response jointly define postoperative risk, then patients at the high-risk end—those with non-major responses and lingering nodal disease—might be candidates for intensified adjuvant therapy, closer surveillance imaging, or enrollment in trials of additional systemic treatment. Conversely, patients with major pathological responses and complete nodal clearance, whose three-year disease-free survival approaches 90 percent, might eventually be spared some of the toxicity of aggressive adjuvant regimens. The framework could also refine how clinical trials stratify patients after neoadjuvant immunochemotherapy, replacing coarse nodal categories with a graded measure of residual burden that better reflects biological reality.

Caveats remain, and the study acknowledges them plainly. As a retrospective cohort analysis, it cannot exclude residual confounding, and the 286-patient sample, while adequate for the models tested, is modest by the standards of prognostic research. The nodal burden measure also depends on the quality of lymphadenectomy—the number of nodes examined and removed—which can vary between surgeons and institutions. Most importantly, the authors state that prospective validation is required before clinical implementation. Still, the study lands at a propitious moment: immunochemotherapy is rapidly becoming standard for locally advanced gastric cancer, and the field urgently needs postoperative tools that match the sophistication of preoperative treatment. By showing that the arithmetic of residual lymph nodes—counted one doubling at a time—carries independent prognostic weight, this research offers a concrete, reproducible step toward truly individualized risk assessment after surgery.

Subject of Research: Risk stratification after neoadjuvant immunochemotherapy in locally advanced gastric cancer using pathological response and residual nodal burden

Article Title: Primary-tumor pathological response and postoperative residual nodal burden for risk stratification after neoadjuvant immunochemotherapy in gastric cancer

Article References: Yu, M., Qi, Y., Huang, G., Zhang, J., Li, C., & Li, Z. (2026). Primary-tumor pathological response and postoperative residual nodal burden for risk stratification after neoadjuvant immunochemotherapy in gastric cancer. BMC Cancer. https://doi.org/10.1186/s12885-026-16953-9

Image Credits: AI Generated

DOI: 10.1186/s12885-026-16953-9

Keywords: gastric cancer, neoadjuvant immunochemotherapy, pathological response, residual lymph-node burden, risk stratification, disease-free survival, gastrectomy, gastroesophageal junction adenocarcinoma, Cox regression, C-index, cancer staging, prognosis