A multistage stratified cluster random sampling was employed to assess the nutritional status of hospitalized patients in China from August 2020 to August 2021.

The sample size was determined based on the overall incidence of nutritional risk among 6,638 hospitalized patients in 34 hospitals in China in 2017, which was 42.34% (25). The sample size for each stratum was calculated using the following formula.

The acceptable error range, denoted E, was established at ±1% (E = 0.01). The standard deviation corresponding to a 95% confidence level, denoted Z, was approximately rounded to 1.96 (Z = 2). The empirical statistic for the total incidence of nutritional risk in hospitalized patients, denoted as P, was 0.4234. By substituting these values into the formula, the sample size could be calculated as N = 9,765. Taking into account a follow-up loss rate of 20%, we estimated that each stratum would require a total of 11,718 individuals. This survey was stratified based on three regions: eastern, central, and western; therefore, a total of 35,154 patients were required for this study.

The inclusion criteria were as follows: (1) patients newly admitted to the hospital with diseases in any of the seven specialty systems; (2) age ≥ 18 years old; and (3) within 24–48 h after admission.

The exclusion criteria encompassed the following: (1) pediatric and critically ill patients; (2) individuals with mental illness or memory disorders that impeded accurate response to questions; (3) those lacking behavioral capacity; and (4) other conditions deemed unsuitable for inclusion by the investigators.

The calculation formula for SII, which is based on peripheral neutrophil count (N), lymphocyte count (L), and platelet count (P), can be mathematically expressed as SII = N × P/L. Additionally, the SII was transformed into the natural logarithm (lnSII) and stratified by the interquartile range (IQR) to meet the data analysis requirements in our study.

The diagnosis of malnutrition in this study consisted of two primary steps: nutritional risk screening and malnutrition assessment. Figure 1 presents a diagnostic flow chart that illustrates the identification process. Initially, nutritional risk screening was followed by a comprehensive evaluation using the Malnutrition Screening Tool 2002 (NRS 2002) to identify patients at risk (score ≥ 3) (26). For details on the NRS 2002 assessment and results, please refer to Supplementary Tables S1, S2. Subsequently, an assessment was performed based on the Global Leaders Initiative on Malnutrition (GLIM) criteria, requiring patients to meet both phenotypic and etiologic indicators for a confirmed diagnosis (11). Supplementary Table S3 shows the detailed diagnosis of malnutrition.

To validate the efficacy of SII compared to other nutritional screening tools, we introduced the controlling nutritional status (CONUT) score in this study. The CONUT score is an additional screening tool that can objectively and simply assess the nutritional status of hospitalized patients. It comprises three indicators: serum albumin (ALB), total cholesterol (TC), and peripheral lymphocyte count. Among them, albumin reflects protein reserves; total cholesterol is associated with caloric deficiency, and lymphocyte count is related to immune defense (27). Unlike the NRS 2002 + GLIM criteria diagnostic model, the CONUT score does not take into account anthropometric measures that may be influenced by effusion and edema, such as body mass index (BMI), upper arm circumference, and calf circumference (28). The scoring criteria are determined based on the ranges of the three indicators, and the CONUT score is the sum of the three scores. As presented in Table 1, four groups of patients with varying degrees of undernutrition were defined based on the CONUT score: normal (0–1), mild (2–4), moderate (5–8), and severe (9–12).

Furthermore, our study incorporated covariates that may influence outcomes. Sociodemographic factors included gender, age, and education levels. Physical examination included BMI, waist circumference, hip circumference, upper arm circumference, calf circumference, and grip strength. Blood test indices comprised white blood cell count (WBC), red blood cell count (RBC), hemoglobin (Hb), total cholesterol (TC), triglyceride (TG), and albumin (ALB). Measurement methods of physical examination indicators and normal reference value range of blood test indices can be found in Supplementary Tables S4, S5.

Data imputations and analyses were performed using IBM SPSS Statistics (version 27.0) and RStudio (version 4.3.0), utilizing packages such as “rms” for regression modeling, “ggplot2” for graphic visualization, and “pROC” for ROC curve analysis.

Multiple imputation was employed to address missing values in this study. By comparing the data before and after imputation, the optimal dataset was selected for handling missing values. Categorical variables were described using rates or composition ratios (%), and comparisons between groups for unordered variables were made using the chi-square test. Continuous variables following a normal distribution were described as mean ± standard deviation (SD) and compared using the t-test. For non-normally distributed continuous variables, the median (P₂₅–P₇₅) was used, and comparisons were made using the Mann–Whitney U test. All tests were two-tailed, with a significance level of α = 0.05.

A series of stepwise logistic regression models were conducted to explore the relationship between SII and malnutrition. The crude model did not incorporate any covariates. Model I included sociodemographic characteristics. Model II further adjusted for physical examination characteristics, and Model III additionally accounted for disease characteristics.

Restricted cubic spline (RCS) was employed to visually represent the dose–response relationship between SII and malnutrition risk, elucidating potential nonlinearity. To avoid overfitting with RCS, linear spline was also used to simulate their linear association. This comprehensive approach facilitated an accurate assessment of the link between SII and malnutrition.

To assess the suitability of SII in the CONUT score, multiple logistic regression was used to explore potential risk factors associated with varying degrees of malnutrition. SII was categorized into quartiles, with the first quartile (Q₁) as the reference. Normal nutritional status served as the reference for the dependent variable. Models were iteratively adjusted, odds ratios (ORs) were calculated, and corresponding forest plots were generated.

In the consistency analysis, we validated the effectiveness of two methods for diagnosing malnutrition by calculating the Kappa value and analyzing consistency data. A Kappa value of 0.21–0.40 indicated fair agreement, 0.41–0.60 moderate agreement, and 0.61–0.80 substantial agreement. The predictive ability of SII within the consistency data was compared using receiver operating characteristic (ROC) analysis, and the area under the curve (AUC) was calculated.

The baseline characteristics of the subjects are provided in Table 2. A total of 40,379 participants were included in this study. Among them, 56.7% were men, with a mean age of 59.6 ± 14.3 years, and more than half had primary or middle school education. Circulatory diseases had the highest prevalence among the seven system diseases, accounting for 36.1%, while urogenital diseases had the lowest prevalence at 15.2%. The diagnosis of malnutrition was established using the NRS 2002 + GLIM criteria, identifying a total of 5,400 patients with malnutrition, resulting in a prevalence of approximately 13.4%. Individuals with malnutrition exhibited a higher likelihood of being male, having less than a junior high school education, and comorbidities such as tumors, respiratory diseases, and digestive diseases. Additionally, malnourished patients showed higher levels of SII, age, and WBC; while demonstrating lower BMI, waist circumference, hip circumference, upper arm circumference, calf circumference, and grip strength; as well as reduced levels of RBC and Hb along with decreased TC, TG, and ALB.

Furthermore, Table 2 presents the results of the t-tests and the chi-square tests conducted for each variable. All covariates, with the exception of TC, exhibited a significant association with malnutrition (p < 0.05). Among the variables under investigation, lnSII was found to be significantly higher in the malnutrition group compared to the non-malnutrition group (6.28 ± 0.78 vs. 6.63 ± 0.97, p < 0.001). Furthermore, a clear trend emerged in the prevalence of malnutrition when converting SII to IQR. Initially, there was a downward trend, but this was followed by an increase as SII rose from Q₂ to Q₄. Notably, the lowest prevalence was observed at 8.9% when SII was in Q₂.

To further validate the association between SII and the risk of malnutrition, a stepwise binary logistic regression model was constructed, as delineated in Table 3. In the crude model, participants in Q₂ (OR = 0.91, 95%CI: 0.82–0.99) exhibited a significantly lower risk of malnutrition compared to those in Q₁. However, with the elevation of SII to Q₃ and Q₄, there was a substantial increase in risk with corresponding ORs of 1.36 (95%CI: 1.25–1.49) and 2.58 (95%CI: 2.39–2.80), respectively. The results from Model I and Model II remained robust and statistically significant even after adjusting for multiple variables through multivariate analysis. However, with the introduction of disease characteristic variables in Model III, the statistical significance in Q₂ was no longer evident (OR = 0.94, 95% CI: 0.85–1.05). Nevertheless, the general associations in the four models remained largely unchanged, indicating a correlation between higher levels of SII and an increased risk of malnutrition.

As illustrated in Figure 2, all four models demonstrated a significant nonlinear association between SII and malnutrition (P for nonlinear <0.001). In the crude model, the RCS depicted a negative relationship between lnSII and malnutrition until it reached the lowest point at a lnSII level of 5.77 (OR = 0.79, 95%CI = 0.72–0.86); thereafter, the gradual increase in the level of lnSII was associated with an increased risk of malnutrition, revealing a U-shaped dose–response pattern. A similar trend was also evident in the adjusted model, where the OR value reached its minimum around a lnSII of 5.75 (corresponding to SII = 315 × 10⁹ cells/μL). However, linear spline fitting in all four figures consistently indicated a continuous rise in the risk of malnutrition as lnSII increased.

Considering the potential interactions between different types of diseases, we further analyzed the cut-off values of SII in cases where malnutrition was present and only one type of disease was involved. Specifically, for seven different types of diseases, the cut-off values of SII in the presence of malnutrition were presented in Figure 3. Among these, the cut-off value for respiratory diseases (SII = 1284.4 × 10⁹ cells/μL) is the highest, indicating that patients with respiratory diseases have the strongest inflammatory response in the presence of malnutrition. Conversely, the cut-off value for neurological disorders (SII = 488.5 × 10⁹ cells/μL) is the lowest, suggesting that malnourished patients with neurological disorders have a comparatively lower inflammatory response. These results suggested that the cut-off values of SII vary significantly among patients with different diseases when malnutrition is present. Such differences reflected the varying impacts of malnutrition on the inflammatory response in different disease states. The ROC results for the seven diseases were presented in Supplementary Figure S1.

The descriptive characteristics of the 40,379 participants based on their CONUT scores were presented in Table 4 (The full table was provided in Supplementary Table S6). In general, the prevalence of patients with normal and mild malnutrition was 40.1 and 45.0%, respectively, with the remaining 14.9% displaying moderate and severe malnutrition. It could be observed that among participants without malnutrition, there was a decrease in the prevalence of normal individuals from 51.9% to 17.6% as SII increased from Q₁ to Q₄. On the contrary, contrasting results were evident within the malnutrition population, where an increase in SII corresponded to an overall rise in malnutrition prevalence, indicating a positive correlation trend. Interestingly, a U-shaped trend in prevalence was also observed in patients with moderate and severe malnutrition compared to those who were normal or mildly malnourished. Likewise, the prevalence reached its minimum when SII was in Q₂.

We used forest plots to visualize multiple logistic regression results for different levels of malnutrition (Figure 4). In mild malnutrition, elevated SII was a risk factor in all four models before and after adjustment (OR > 1). Although among moderate malnutrition cases, SII within Q₂ exhibited a potential protective factor compared to Q₁ (OR < 1). Similar findings were observed in cases of severe malnutrition. However, the adjusted model revealed that when SII was in Q₃, it no longer exhibited statistical significance in severe malnutrition, while there was still a tendency for an increased risk of malnutrition observed in Q₄. The specific OR could be found in Supplementary Table S7.

To validate the diagnostic effectiveness of the two methods, patients with normal and mild malnutrition on the CONUT score were combined into a group, while those with moderate and severe malnutrition were grouped, ensuring consistency with the grouping basis of the NRS 2002 + GLIM criteria (29). The results of the Kappa consistency test (Kappa = 0.23) revealed a low level of agreement between the severity of malnutrition determined by the NRS 2002 + GLIM criteria and the CONUT score (Figure 5A). However, SII performed well in both methods, indicating its effectiveness in prediction. Finally, the consistency data obtained from both methods were combined and a ROC analysis was conducted on the merged dataset (Figure 5B). The resulting area under the curve was 0.73 (95%CI: 0.714–0.742), indicating the stability of our findings.