Overweight BMI Linked to Lower Mortality in Gastric Cancer

A higher body mass index in the overweight range was associated with lower mortality among adults with gastric cancer in a retrospective observational cohort from Fujian Cancer Hospital. Among 374 patients included in the primary multivariable analysis, those with a baseline BMI of 24.0–27.9 kilograms per square metre had a 38% lower mortality hazard than patients with a BMI of 18.5–23.9 after adjustment for clinical, nutritional, inflammatory, functional and treatment-related factors. The hazard ratio was 0.62, with a 95% confidence interval of 0.41–0.93 and a p-value of 0.020. The finding applied to the overweight category rather than to obesity: the 12 patients with a BMI of at least 28.0 had an adjusted hazard ratio of 0.51, but its wide confidence interval of 0.20–1.30 included no effect, and the estimate was not statistically significant.

The study included 380 adults newly diagnosed with histologically confirmed gastric cancer between September 1995 and May 2017. Baseline was defined as the date of first pathological diagnosis, when researchers collected demographic, tumour, treatment, laboratory, nutritional and functional information. The participants had a mean age of 59.5 years, and 65, or 17.1%, were underweight, 220, or 57.9%, were in the normal-weight range, 83, or 21.8%, were overweight, and 12, or 3.2%, were obese. Overall survival was the interval from first histopathological diagnosis to death from any cause, with patients alive at their last documented contact censored. Follow-up ranged from 0.2 to 67.8 months, and the reverse Kaplan–Meier estimate of median follow-up was 37.0 months.

Nutritional and functional differences were pronounced across the BMI categories. Every patient classified as underweight had a Nutritional Risk Screening 2002 score of at least 3, compared with 64.1% of normal-weight patients, 63.9% of overweight patients and 50.0% of obese patients. Severe nutritional impairment, represented by a Patient-Generated Subjective Global Assessment score of at least 9, occurred in 55.4% of underweight patients, 35.5% of normal-weight patients, 28.9% of overweight patients and 8.3% of obese patients. Median Karnofsky Performance Status was 80 in the underweight group and 90 in each of the other groups. Prealbumin concentrations also increased across the categories, while radical surgery was more common among underweight patients. These differences help explain why BMI cannot be interpreted as a biological cause of better survival: it may instead act as a marker of nutritional reserve, disease-related weight loss, functional status and treatment selection.

Kaplan–Meier analysis showed an overall difference in survival among the four BMI categories, with a log-rank p-value of 0.011, and unadjusted survival was lowest in the underweight group. The multivariable model included BMI category, age modelled with a restricted cubic spline, clinical TNM stage, PG-SGA category, Karnofsky Performance Status, prealbumin, neutrophil-to-lymphocyte ratio and baseline chemotherapy, with baseline hazards stratified by surgical procedure and nutritional support. In that model, underweight was not independently associated with overall survival, with a hazard ratio of 0.89 and 95% confidence interval of 0.60–1.34. Advanced disease remained strongly associated with mortality: compared with stage I, the hazard ratio was 4.63 for stage III and 10.93 for stage IV. A 10 mg/L increase in prealbumin was associated with a lower adjusted hazard of 0.972, while receipt of chemotherapy at baseline was associated with a lower hazard of 0.653. The latter variable represented treatment already initiated at baseline and did not specifically identify postoperative adjuvant chemotherapy.

Stage-specific sensitivity analyses suggested different patterns in early and advanced disease, but the formal test did not establish that BMI modified the survival association by stage. In 89 patients with stage I–II disease, including 26 deaths, underweight was associated with a higher mortality hazard than normal weight, with a hazard ratio of 5.19 and a 95% confidence interval of 1.42–18.96. The interval was wide, reflecting the small number of deaths, and neither the overweight nor obesity estimates in the early-stage group were statistically significant. Among 285 patients with stage III–IV disease, including 176 deaths, overweight was associated with lower mortality, with a hazard ratio of 0.61 and 95% confidence interval of 0.41–0.93. Underweight and obesity were not significantly associated with mortality in that advanced-stage subgroup. However, the categorical BMI-by-stage interaction test was not significant, with p=0.173, so the apparently contrasting subgroup results should not be treated as proof that BMI operates differently at each disease stage.

Other nutritional tools also provided partly overlapping information. PG-SGA categories significantly distinguished unadjusted overall survival, with a log-rank p-value of 0.0097, and a score of at least 9 was associated with a higher mortality hazard than a score of 0–3 in the univariable analysis, with a hazard ratio of 1.65. That association was no longer significant after adjustment for stage, BMI, laboratory measures, functional status and treatment variables. NRS2002, by contrast, did not distinguish survival in Kaplan–Meier analysis, with a log-rank p-value of 0.80, and its nutritional-risk category was not associated with mortality in the univariable Cox model. The two instruments were designed for different purposes: NRS2002 screens hospitalized patients for nutritional risk, whereas PG-SGA assesses cancer-related nutrition-impact symptoms, intake, weight change and functional status. In this cohort, about 70% of patients had NRS2002 scores of at least 3, which may have limited its ability to separate patients with different long-term prognoses.

Laboratory and functional measures similarly appeared prognostic before adjustment but were attenuated once disease severity and other correlated factors were included. Better Karnofsky Performance Status, higher albumin and prealbumin, and a lower neutrophil-to-lymphocyte ratio were each associated with lower mortality in univariable analysis, whereas higher neutrophil-to-lymphocyte ratio was associated with increased mortality. In the multivariable model, PG-SGA, Karnofsky Performance Status and neutrophil-to-lymphocyte ratio were not independently associated with overall survival, while prealbumin retained an association. This attenuation is consistent with substantial overlap among nutrition, inflammation, tumour burden and functional reserve. Serum protein concentrations can reflect inflammation, hepatic synthesis and fluid status as well as nutrient intake, and performance status can influence both treatment selection and survival. Restricted cubic spline analysis detected no significant overall or nonlinear association between continuous BMI and mortality, with an overall p-value of 0.640 and a nonlinearity p-value of 0.739.

The investigators also built an exploratory model to estimate survival at one, three and five years. A clinical reference model included age, sex, TNM stage and baseline chemotherapy, while an extended model added continuous BMI, PG-SGA score, Karnofsky Performance Status, prealbumin and neutrophil-to-lymphocyte ratio. The extended model fitted the data better than the clinical model, with a likelihood-ratio chi-square value of 21.80 on five degrees of freedom and p<0.001. After 1,000 bootstrap resamples, the optimism-corrected concordance index rose from 0.646 for the clinical model to 0.681 for the extended model. Time-dependent areas under the curve were 0.744, 0.748 and 0.702 for the extended model at one, three and five years, compared with 0.689, 0.715 and 0.684 for the clinical model. Calibration was generally acceptable at one and three years but more variable at five years. These internal validation results indicate modest incremental discrimination, not proven clinical utility, because the model was not tested in an independent cohort.

Several limitations constrain interpretation of the observational findings and prevent a causal conclusion. The long recruitment period may have included changes in staging, surgery and systemic treatment, and residual confounding from tumour burden, inflammation, comorbidity and treatment selection cannot be excluded. BMI was measured once and cannot distinguish skeletal muscle from adipose tissue, identify visceral adiposity or sarcopenic obesity, or capture recent involuntary weight loss. The obesity group contained only 12 patients, and the early-stage subgroup had just 26 deaths. C-reactive protein, histological differentiation and six-month weight-loss information were available for subsets only; in the 43 patients with six-month weight-loss data, losing at least 5% was associated with higher mortality, but the estimate was exploratory. The prognostic model also requires external validation. The evidence therefore supports BMI of 24.0–27.9 as a prognostic marker associated with lower mortality in this population, while showing that a broader assessment incorporating nutrition, function and inflammation may add limited predictive information. It does not establish that increasing body weight or BMI improves survival.

Subject of Research: Associations between baseline BMI, nutritional status and overall survival in adults with gastric cancer

Article Title: Body mass index and overall survival in patients with gastric cancer: stage-specific associations and prognostic implications

Article References: Zhou, S., Yu, D., Wang, X., Zhan, Z., Guo, Z., & Yu, J. (2026). Body mass index and overall survival in patients with gastric cancer: stage-specific associations and prognostic implications. Frontiers in Nutrition, 13, Article 1937418. https://doi.org/10.3389/fnut.2026.1937418

Image Credits: AI Generated

DOI: 10.3389/fnut.2026.1937418

Keywords: gastric cancer, body mass index, overall survival, nutritional status, PG-SGA, prognostic modeling, observational cohort, mortality risk, Body, mass, index, overall