The National Health and Nutrition Examination Survey (NHANES) database, accessible at (https://www.cdc.gov/nchs/nhanes/index.htm), provides extensive coverage through various indicators. It enables the retrieval of detailed demographic statistics, comprehensive socioeconomic information, dietary and health data, physiological metrics, laboratory test results, and other related data, encompassing a broad cross-section of the United States. Administered by the Centers for Disease Control and Prevention (CDC) and the National Center for Health Statistics (NCHS), NHANES aims to provide comprehensive, nationally representative data on the health and nutritional status of the civilian population in the United States. Before participating, all subjects provided informed consent, and the NCHS Ethics Review Board approved the data collection methodologies used in NHANES.

This investigation included 45,566 individuals who participated in nine consecutive NHANES survey cycles from 2001 to 2018. Individuals under 40 (n=15,766) and those without a cancer diagnosis or known survival status (n=25,056) were excluded from the study. Additionally, we excluded participants lacking GNRI-related data or those with incomplete covariates (n=1,491). After applying the above criteria, 3,253 cancer survivors were available for analysis. Figure 1 depicts a flowchart illustrating the criteria employed for patient selection.

The GNRI (Geriatric Nutritional Risk Index) is a validated, straightforward scoring system developed specifically to assess the nutritional status of patients across various medical settings, notably among those undergoing surgical procedures. This index is highly precise, relying solely on objective parameters such as serum albumin levels, height, and weight, which are readily available from standard laboratory tests. The formula used to calculate the GNRI is: GNRI = (1.489 × serum albumin level in g/L) + (41.7 × actual body weight/ideal body weight in kg), where a ratio of actual to ideal weight set at 1 is assumed. This method has been recognized as a significant predictor of outcomes in diverse cancer populations, illustrating its broad applicability and reliability (11). For practical application, GNRI scores lead to the classification of patients into two primary groups: those with a GNRI < 98 form the Low-GNRI group indicating higher nutritional risk, and those with a GNRI ≥ 98 compose the High-GNRI group suggesting lower nutritional risk (12–14). To refine the understanding of nutritional risk gradients, GNRI scores were further segmented into three tertiles. This categorization helps clinicians identify patients at varying degrees of malnutrition risk more distinctly, with Tertile 1 representing the highest risk and Tertile 3 the lowest, thus enabling more targeted nutritional interventions based on the severity of risk (14).

The mortality data for this analysis were obtained from the NHANES Public Use Linked Mortality File and integrated with standard NHANES datasets using the unique respondent sequence number assigned to each participant. The analysis encompassed mortality status and follow-up duration, classifying outcomes into two categories: survival and death. Cancer mortality is defined as the likelihood of dying from various malignant tumors, whereas non-cancer mortality refers to the likelihood of death from causes other than cancer. The National Death Index (NDI) records were used to determine the mortality status and causes of death until 31 December 2019. In the 10th edition of the International Classification of Diseases (ICD-10), it is outlined that deaths attributed to malignant neoplasms are categorized as cancer mortality (with the specific codes of ICD-10 C00-C97) (15).

Patients with a previous diagnosis of cancer were identified through the inquiry: “Have you ever been told by a doctor that you had cancer or a malignancy of any kind?” Individuals responding “YES” were selected for inclusion. Following this selection, cancer patients were further queried regarding the specific type of cancer diagnosed.

Data regarding the demographic characteristics of the patients, including age, gender, race, education level, and marital status, was collected. Additional covariates were derived from physical examinations conducted at a mobile examination center and from laboratory test results. Ethnicity was divided into four distinct groups: Mexican American, Non-Hispanic White, Non-Hispanic Black, and Other. Education levels were dichotomized as “Below High School” and “High School or Above.” Marital status was categorized into four groups: Married/Cohabitant, Widowed, Separated/Divorced, and Never Married. Body Mass Index (BMI) was determined by dividing weight in kilograms (kg) by the square of height in meters (m²). Based on these calculations, BMI values were categorized into three groups: Underweight or Normal (<25.0 kg/m²), Overweight (25.0 to 30.0 kg/m²), and Obesity (≥30.0kg/m²) (16). Physical activity was classified as either moderate/vigorous recreational activities (yes) or none. Additionally, the energy intake was assessed using dietary data obtained from 24-hour dietary recalls. Smoking status was classified into three categories: Never Smoker, Former Smoker, and Current Smoker. A “Never Smoker” is defined as an individual who has smoked fewer than 100 cigarettes in their lifetime. A “Former Smoker” refers to someone who has smoked more than 100 cigarettes in their lifetime but has quit smoking. A “Current Smoker” is defined as someone who has smoked more than 100 cigarettes in their lifetime and continues to smoke occasionally or daily. Alcohol consumption was categorized based on the past year’s responses into five groups: Never (fewer than 12 drinks in a lifetime), Mild (≤ 2 drinks every day for males, ≤ 1 drink every day for females), Moderate (≤ 3 and > 2 drinks every day for males, ≤ 2 and > 1 drink every day for females, or ≥ 2 to < 5 days per month of binge drinking), Heavy (≥ 4 drinks every day for males, ≥ 3 drinks every day for females, or ≥ 5 days per month of binge drinking), and Former (no drinking in the past year but over 12 drinks in any previous year). The study identified chronic diseases, including diabetes, hypertension, coronary heart disease (CHD), and stroke. Diabetes was further classified into diabetes mellitus (DM), compromised fasting glycemia, and impaired glucose tolerance. Diagnosis of these conditions relied on patient self-reporting.

In this study, we rigorously adhered to the NHANES complex survey design principles, taking into account weights, clustering, and stratification to ensure representative and accurate analyses. We initiated our analysis by categorizing cancer survivors based on their Geriatric Nutritional Risk Index (GNRI) scores into two groups: Low GNRI (< 98) and High GNRI (≥98). For the descriptive analysis, we employed the weighted mean ± standard deviation to summarize continuous variables, ensuring that the complex survey design was accounted for in these estimations. Categorical variables were reported as frequencies and percentages, with survey design adjustments applied to accurately reflect the NHANES sample population. This method enabled a comprehensive comparison of baseline characteristics between the Low GNRI and High GNRI groups, utilizing χ2 tests for categorical variables and ANOVA for continuous variables.

Next, in the analysis of the relationship between GNRI and mortality risk among cancer patients, Cox proportional hazards regression models were employed. The Cox regression model was used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs). The proportional hazards assumption was tested using the Schoenfeld residuals method. These models facilitated the examination of GNRI both as a continuous variable, to assess the impact of each unit change in GNRI on mortality risk, and as a categorical variable, to discern potential nonlinear relationships and risk distribution. Model 1 incorporated age, ethnicity, and gender as variables and accounted for them through modifications. Model 2 expanded upon the changes made in Model 1 by including additional factors such as BMI, marital status, education level, drinking status, smoking behaviors, and the Poverty Income Ratio (PIR). Model 3 extended the range of adjustments to encompass energy intake, physical activity, hypertension, coronary heart disease (CHD), diabetes, stroke, and tumor types, in addition to those already considered in Model 2. This approach allowed us to reveal subtle trends and potential risk thresholds, while thoroughly accounting for a wide spectrum of potential confounders that could influence this relationship.

Kaplan-Meier curves were generated to illustrate the survival probabilities for the three groups based on GNRI status, depicting all-cause mortality, cancer-specific mortality, and non-cancer mortality. Survival distributions across these categories were compared using the log-rank test. Additionally, a restricted cubic spline (RCS) regression with four knots was conducted to assess the non-linear relationships between GNRI and mortality risk. Upon detecting nonlinearity, two-piecewise Cox proportional hazards regression models were constructed, centered around the inflection point. In the subgroup analysis, we further explored the relationship between GNRI scores and mortality risk across various demographic and clinical subgroups. This process entailed stratifying the sample based on factors such as age, gender, ethnicity, and comorbid conditions, and evaluating the association between GNRI scores and mortality within each stratum. We utilized interaction tests to determine if the effect of GNRI on mortality varied significantly across these subgroups, as presented in forest plots. The P-interaction values were derived from Cox proportional hazards regression models, which were adjusted for the complex survey design to ensure that our findings accurately reflected the diversity of the NHANES population. Statistical analyses were performed with version 4.3.2 of R software (R Foundation for Statistical Computing). Weighted regression analysis utilized the “survey” package; construction of the RCS regression harnessed the “rms” package. In this investigation, a two-sided P-value of <0.05 was considered to be statistically significant.

A total of 3,253 cancer survivors were included in this study, representing a weighted population of 15,723,705 individuals. The weighted mean (standard error) age was 64.78 (0.30) years, with 44.1% male and 55.9% female. Table 1 presents a comparison of the characteristics between the High-GNRI group (GNRI ≥ 98, n = 2963) and the Low-GNRI group (GNRI < 98, n = 290). The majority of patients were non-Hispanic White (88.38%). The p-values indicated significant differences in ethnicity, education level, PIR, BMI, smoking status, diabetes, and tumor types among the groups (P < 0.05). However, age, gender, marital status, drinking status, energy intake, physical activity, hypertension, stroke, and CHD did not exhibit significant differences between the two groups concerning the nutritional risk indicated by the GNRI score (P > 0.05). Significantly, the high GNRI group (GNRI ≥ 98) primarily comprised non-Hispanic White individuals who possessed at least a high school education and consumed more energy. These individuals were either former or never smokers, were identified as mild drinkers, had a higher BMI, and did not have diabetes. Additionally, a higher proportion of participants with gastrointestinal (5.94%) skin (38.48%) and gynecologic (37.05%) cancers were included.

Figure 2 depicts Kaplan-Meier survival curves for all-cause, cancer, and non-cancer mortality of survival among American adult cancer patients stratified by the GNRI group ( Figures 2A–C ). The survival analysis elucidated significant differences in survival probabilities between these cohorts. Notably, the cohort with GNRI scores < 98 exhibited substantially reduced survival rates across all evaluated mortality domains, compared to their counterparts in the ≥98 GNRI cohort. This discrepancy in survival outcomes was statistically significant, as evidenced by the log-rank test results, which yielded p-values of <0.001 for all-cause, cancer, and non-cancer mortality.

After adjusting for multiple covariates, a multivariate analysis was conducted using three distinct regression models to investigate the predictive value of GNRI on overall mortality, cancer-related mortality, and non-cancer-related mortality. The GNRI is assessed both as a continuous variable and in categorized forms. The findings are displayed as hazard ratios (HRs) alongside their corresponding 95% confidence intervals (CIs) ( Table 2 ).

Across all models, a higher GNRI, treated as a continuous variable, was associated with a statistically significant reduction in the risk of all-cause mortality (HR < 1, P < 0.001), cancer mortality (HR < 1, P < 0.001), and non-cancer mortality (HR < 1, P < 0.001). This indicates that higher GNRI scores are protective against mortality. In categorical analyses, individuals with a GNRI ≥98 had significantly lower hazard ratios (HRs) for all-cause, cancer-specific, and non-cancer mortality compared to the reference group (GNRI < 98), with HRs of 0.49(95% CI: 0.38,0.62),0.34(95% CI: 0.23,0.50), and 0.60(95% CI: 0.45,0.79) in Model 3, respectively.

The stratification of GNRI resulted in the identification of three distinct categories: Tertile 1 (33.70, 112.97], Tertile 2 (112.97, 123.12], and Tertile 3 (123.12, 203.79]. Tertile 1 served as the referential baseline. After adjusting for potential confounding variables, observations across the three models unveiled a consistent trend. Specifically, in the most adjusted Model 3, the HRs for all-cause, cancer-specific, and non-cancer mortality among individuals in Tertile 3, compared with Tertile 1, were 0.51 (95% CI: 0.39–0.66), 0.37 (95% CI: 0.23–0.59), and 0.53 (95% CI: 0.38–0.75), respectively. The significance of this trend was supported by P-values for the trend, which were <0.001 for all mortality outcomes in every model, thereby strengthening the correlation between GNRI scores and risk of mortality.

In a fully adjusted restricted cubic spline regression model (RCS), we detected a significant nonlinear relationship between GNRI and all-cause, cancer-specific, and non-cancer mortality among cancer patients. This was evidenced by the P-Nonlinear values of <0.001 for all-cause mortality, 0.024 for cancer-specific mortality, and <0.001 for non-cancer mortality ( Figure 3 ). The RCS analysis indicates that patients with extremely low or high GNRI scores are significantly associated with an increased risk of mortality for all-cause, cancer-specific, and non-cancer mortality. It also shows that as the GNRI increases to a certain point, mortality risk decreases before slightly increasing again. This suggests that maintaining a GNRI score around this optimal value may benefit patients, scores outside this range are associated with a higher mortality risk.

After identifying a non-linear relationship, we utilized two segmented linear regression models to delineate the changes in the relationship between GNRI and mortality risk at specific threshold points ( Table 3 ). Thresholds were determined by evaluating all possible values and selecting the points with the highest log-likelihood values. The established thresholds were 134.16 for all-cause mortality, 134.65 for cancer mortality, and 134.15 for non-cancer mortality, indicating significant changes in the linear relationship between GNRI and mortality risk. For all-cause mortality, when GNRI was below the threshold of 134.16, each unit increase resulted in a slight but statistically significant reduction in mortality risk, with a hazard ratio (HR) of 0.99 (95% CI: 0.98, 0.99). Above this threshold, the HR was 1.00 (95% CI: 0.98, 1.02), suggesting that further increases in GNRI did not significantly alter mortality risk. In the models for cancer mortality and non-cancer mortality, the HR below the thresholds also indicated a statistically significant risk reduction.

The P-values for the log-likelihood ratio tests were statistically significant across all outcomes (all-cause and non-cancer mortality <0.001, cancer mortality 0.024), confirming that the segmented linear models were more suitable for these data than the basic linear models and that there is indeed a non-linear association between GNRI and various forms of mortality.