Data were obtained from the China Nutrition Fundamental Data 2020 (CNFD 2020) Project (14), which was conducted from February to October 2023 in 31 provinces, municipalities and autonomous regions in China using a multi-stage stratified cluster sampling method based on China's administrative divisions. In the first stage, 31 provinces, municipalities and autonomous regions were selected in China. The second stage involved selecting 1–23 tertiary hospitals and 1–2 secondary hospitals with nutrition departments from each province, with one provincial key hospital designated as the coordinating center. In the third stage, continuous sampling was conducted at each site until the predetermined target sample size (200 patients per tertiary hospital and 150 patients per secondary hospital) was reached.

Participants were included in the survey if they were aged 18 years or over, had been hospitalized for tumors or diseases of the respiratory, digestive, endocrine, cardiovascular, urinary or nervous systems, and had completed the data collection within 24–48 h of admission. In this study, the inclusion criteria were further refined to include only participants with a primary diagnosis of neurological disease from the surveyed population. The exclusion criteria included pediatric and critically ill patients, individuals with mental disorders or memory impairment who could not provide accurate responses, participants who lacked decision-making capacity, and other patients deemed ineligible by the investigators. In this study, those who did not undergo malnutrition risk screening were further excluded.

In terms of data collection, dietitians and clinicians conducted face-to-face interviews with patients to gather sociodemographic data; anthropometric measurements were performed according to standard protocols; laboratory data were extracted from the hospital information system; and body composition was measured using BIA (for some patients).

This study strictly adhered to the guidelines of the STROBE Statement and received ethical approval from the Ethics Committee of Peking Union Medical College Hospital (I-22PJ744), with all participants providing verbal informed consent prior to the survey.

The classification of eastern, central, and western regions in this study follows the standard established by the National Bureau of Statistics of China. This classification groups China's 31 provincial-level administrative regions based on geographic location and socioeconomic development levels. Specifically, the eastern region comprises the economically advanced coastal provinces; the central region serves as a major base for grain production and energy raw materials; while the western region is less developed economically and encompasses predominantly inland and remote areas.

The primary hospitalization outcome measures of this study were length of stay and hospitalization costs, with the former defined as the number of days from admission to discharge and the latter including all treatment-related costs incurred during hospitalization.

To ensure consistency, all interviewers and staff members received standardized pre-survey training. Initially, case report forms (CRFs) were completed in paper format and then entered into a secure electronic platform with logical checks by trained interviewers within 7 days. Quality control was implemented at multiple levels: local inspectors at each study site reviewed both paper and electronic CRFs, while a panel of experts randomly reviewed 10% of all CRFs via the electronic platform.

In the data preprocessing stage, we addressed outliers in the measurement data. Values below the 1st percentile or above the 99th percentile for each variable were defined as outliers and replaced with random numbers between the 1st−25th percentile and the 75th–99th percentile, respectively. For missing data (Supplementary Table S1), mean imputation was used to fill in the missing values (20).

For quantitative data that did not meet the assumption of normal distribution, median and interquartile range (IQR) were used for descriptive statistics, and the Wilcoxon rank-sum test was used for comparisons between groups. Variables with statistical significance were screened by univariate analysis and included in a multivariate logistic regression. The specific variables are as follows: KPS, region, WBC, neutrophil percentage, lymphocyte percentage, RBC, hemoglobin, platelet, ALT, potassium, sodium, and chloride. Comparisons between groups were performed as follows: normal nutrition vs. malnutrition, normal nutrition vs. moderate malnutrition, normal nutrition vs. severe malnutrition, and moderate malnutrition vs. severe malnutrition, to analyze the factors influencing malnutrition. Results were presented as odds ratios (OR) with 95% confidence intervals (CI). Additionally, random forest analysis was performed using the same pairwise comparison approach as the logistic regression to model and analyze the predictive factors of malnutrition. In the random forest model, the data were divided into training and test sets in an 8:2 ratio, with synthetic minority oversampling technique (SMOTE) applied to the training set only. The parameters of the random forest model were set as follows: n_estimators =300, max_depth = 10, min_samples_split =10, min_samples_leaf = 4, max_features = “None,” and random_state = 42. The experiment was independently repeated 20 times, and the final results were averaged over the test sets. The final model interpretation was performed using SHapley Additive exPlanations (SHAP) theory to explain the predictions. The Kruskal-Wallis test and multiple linear regression were employed to analyze the associations between malnutrition status and hospitalization costs or length of stay, with results visualized using box plots. A threshold of P < 0.05 was set for statistical significance. All analyses were performed using R (version 4.4.3) and Python (version 3.11.4) (Figure 1).

This study included a total of 1,357 newly hospitalized patients initially diagnosed with neurological diseases from 31 provinces, autonomous regions, and municipalities directly under the central government in China. The results showed that the overall prevalence of malnutrition was 32.5%, with moderate malnutrition accounting for 30.8% and severe malnutrition accounting for 1.7%.

This study included 775 male patients and 582 female patients. The median age (IQR) was 64 (17.0) years. There were 11 patients with a Master's degree or higher, 174 patients with a Bachelor's degree, 293 patients with high school, 735 patients with primary or junior high school, and 144 patients with no formal education. There were 47 patients at dependent level, 294 patients at semi-dependent level, and 1,016 patients at independent level. There were 893 patients (65.8%) with comorbidities. The regional distribution of patients was as follows: 736 patients (54.2%) in the eastern region, 214 patients (15.8%) in the central region, and 407 patients (30.0%) in the western region. Of these patients, 441 were malnourished, with an overall malnutrition rate of 32.5%, including 418 patients with moderate malnutrition (30.8%) and 23 patients with severe malnutrition (1.7%) (Table 1). Laboratory characteristics and physical examination findings are shown in Supplementary Table S2.

The eastern region had a significantly higher proportion of normal patients (55.0%) compared to its malnourished patients (52.6%), while the central region had 163 (17.8%) normal patients compared to 51 (11.6%) malnourished patients. In contrast, the western region demonstrated a significantly higher prevalence of malnutrition (35.8%) compared to its normal population (27.1%). Univariate analysis confirmed a significant association between regional differences and nutritional status (P = 0.042), suggesting that the western region may be a risk area for malnutrition (Table 1, Supplementary Table S2).

Phenotypic results findings: 428 patients (31.5%) with moderate malnutrition had weight loss, 13 patients (1.0%) with severe malnutrition had weight loss; 417 patients (30.7%) with moderate malnutrition had low BMI, 24 patients (1.8%) with severe malnutrition had a low BMI; 412 patients (30.4%) with moderate malnutrition had muscle loss, and 29 patients (2.1%) with severe malnutrition had muscle loss. Compared with other neurological disorders, patients with cerebrovascular disease were the most likely to meet the phenotypic criteria (59.1%). The western region (1.2%, 2.0%, 4.2%) exhibited the highest prevalence of the three malnutrition phenotypes compared to the eastern (1%, 1.8%, 1.5%) and central (0.5%, 1.4%, 0.5%) regions (Table 2).

Etiologic results findings: 342 patients (25.2%) with malnutrition showed reduced food intake/assimilation, and 423 patients (31.7%) with malnutrition showed disease burden/inflammation. The western region (35.6%, 32.4%) showed the highest prevalence of malnutrition meeting both etiologic criteria compared to the central (14.0%, 28.0%) and eastern (22.7%, 31.4%) regions (Table 3).

Based on the results of the univariate analysis (Supplementary Table S3), we excluded non-significant factors (P > 0.05) and entered the remaining variables into logistic regression models, followed by stepwise multivariate logistic regression analyses across different nutritional status groups. In the malnutrition vs. normal group, the higher the KPS score the lower the risk of malnutrition (P < 0.05). The risk of malnutrition was lower in the central region (P = 0.015) and higher in the western region (P = 0.048) compared to the eastern region. The risk of malnutrition was higher with higher neutrophil percentage, lower RBC and chloride (P < 0.05). In moderate malnutrition vs. normal group, KPS score, region, neutrophil percentage and RBC were similar to that of malnutrition vs. normal group. The risk of malnutrition was higher with less chloride (P = 0.011). ALT had no statistical significance. In the severe malnutrition vs. normal group, KPS score, neutrophil percentage and chloride were similar to those in the moderate malnutrition vs. normal group. Lower hemoglobin, platelets and potassium were associated with a higher risk of malnutrition (P < 0.05). In severe vs. moderate malnutrition, KPS score, chloride, hemoglobin and potassium were similar to those in severe malnutrition vs. normal group. Platelets were not statistically significant (Table 4).

Figure 2 displays the top ten most important predictive factors for comparisons between the four malnutrition groups. Figure 2A presents the top 10 predictors for malnutrition, with lymphocyte percentage, RBC, and glucose ranking in the top three, while the top five factors include lymphocyte percentage, RBC, glucose, BUN, and platelet count. Figure 2B shows the top 10 predictors of moderate malnutrition, with lymphocyte percentage, RBC, and glucose as the top three, and the top five being lymphocyte percentage, RBC, glucose, BUN, and hemoglobin. Figure 2C illustrates the top 10 predictors of severe malnutrition, with senior high school education, primary/junior high school education, and BUN ranking highest, followed by senior high school education, primary/junior high school education, BUN, glucose, and RBC in the top five. Figure 2D demonstrates the top 10 factors predicting severe malnutrition among moderate malnutrition patients, with WBC, neutrophil percentage, and primary/junior high school education level being the top three predictors, while the top five consist of WBC, neutrophil percentage, primary/junior high school education, senior high school education, and age 25–35 years.

The developed random forest prediction models performed comparisons between the three malnutrition classification groups and the normal nutrition group, with all models demonstrating excellent predictive performance. Specifically, the models corresponding to Figures 3C, D achieved the highest AUC values of 0.99, followed by Figure 3B (AUC = 0.97), and Figure 3A (AUC = 0.97). Full model performance metrics, including accuracy rates and detailed AUC values for all four models, are provided in Supplementary Table S4.

Figure 4 presents boxplot analyses of the relationship between patient malnutrition status and hospitalization outcomes. As shown in Figure 4A, the median length of stay increased progressively with worsening malnutrition severity, with malnourished patients having longer length of stay than the normal nutrition group, along with more dispersed data distributions and larger IQRs. Figure 4B shows that hospitalization costs increased with the severity of malnutrition. Patients with severe malnutrition had higher hospitalization costs than those with normal nutrition, with costs slightly exceeding those of moderately malnourished patients. Both the Kruskal-Wallis test and multiple linear regression showed no statistically significant difference in malnutrition. OR and 95% CI are shown in Supplementary Table S5.

By comparing the distribution of GLIM indicators across different neurological disease types and regions in China, we found that the western region had significantly higher proportions of severe phenotypic malnutrition indicators and etiologic malnutrition indicators than eastern and central regions. We hypothesize that this disparity may be due to developmental differences in education, economy, and health infrastructure in western regions compared with eastern/central regions, which may exacerbate the severity of malnutrition (2, 25). Furthermore, patients with cerebrovascular disease experienced higher rates of moderate weight loss and low BMI than those with other neurological conditions. This may be due to hypermetabolic states (e.g., spasticity) and chronic inflammation (2, 7). Patients with central nervous system disorders exhibited particularly pronounced moderate to severe muscle mass reduction, suggesting that neurogenic muscle atrophy may represent the core phenotypic manifestation of malnutrition in these disorders.

The high prevalence of malnutrition in neurological patients may be attributed to impaired nutrient intake and disease-specific mechanisms. For instance, patients with stroke or Parkinson's disease often experience loss of arm strength and coordination, leading to difficulties in self-feeding (26, 27). Those with Alzheimer's disease or other dementias may forget to eat due to cognitive decline (28). Additionally, financial constraints and mobility limitations—common in chronic neurological conditions—can further restrict access to adequate nutrition. The use of the KPS helps quantify functional and cognitive impairments, while demographic variables such as education level and geographic region serve as proxies for socioeconomic and environmental barriers that affect both disease management and nutritional access. By integrating functional metrics like the KPS with demographic factors, our study captures this relationship.

Chen et al. used GLIM to investigate the association between malnutrition and neurological disease in elderly patients, with a sample size of 566 patients. Their results showed a malnutrition rate of 14.7%, indicating certain associations between malnutrition and several hospitalization indicators/outcomes (29). Liu et al. (1) applied both GLIM and subjective global assessment (SGA) to investigate the relationship between malnutrition and clinical outcomes in neurocritical patients. This single-center study (n = 161) reported malnutrition rates of 49% (GLIM) and 28% (SGA). The study showed that 22% of malnourished patients met GLIM criteria while only 1% met SGA criteria, demonstrating the superiority of GLIM in diagnosing malnutrition in neurological patients. The single-center study by Zhang et al. (30) (n = 140) used NRS-2002, nutrition risk in the critically ill (NUTRIC) and modified NUTRIC (mNUTRIC) to assess malnutrition in critically ill neurological patients and found nutrition risk rates of 87.1%, 15.7% and 28.6%, respectively. Our study found a malnutrition rate of 32.5% in neurological patients (moderate: 30.8%; severe: 1.7%). We attribute this increased prevalence primarily to disease-specific factors such as dysphagia and reduced food intake due to stroke, dementia and other neurological conditions.

As a multicentre study covering 31 provinces with 1,357 patients using the GLIM criteria, our methodology differed significantly from Liru et al.'s (29) study which used only calf circumference (male ≤ 30 cm, female ≤ 29 cm) to assess muscle mass. We implemented three internationally recognized measures: calf circumference, ASMI, and handgrip strength. These methodological differences in nutritional assessment criteria and indicators are likely to account for the substantial variation in reported malnutrition rates.

Using logistic regression and random forest analyses, we identified neutrophil percentage, lymphocyte percentage, western region, and RBC as significant influencing factors for malnutrition. The study by Kaya et al. demonstrated that the neutrophil-to-lymphocyte ratio (NLR) was an independent risk factor for malnutrition in elderly patients (3). As an inflammatory marker, the NLR reflects the systemic inflammatory status - elevated NLR values (indicating increased neutrophils and decreased lymphocytes) indicate impaired inflammatory homeostasis. Our findings regarding neutrophil and lymphocyte percentages as determinants of malnutrition are consistent with these previous reports. In addition, lower RBC levels indicate a higher likelihood of anemia and an increased risk of malnutrition. Research by Göl et al. (31) suggests a correlation between malnutrition and anemia, which is consistent with the findings of the present study. The consistent prominence of primary/junior high school education level among the top-ranked determinants of different malnutrition statuses suggests that lower educational attainment may impair patients' ability to manage their nutritional needs. Lower educational attainment is often associated with limited financial resources, which may restrict access to adequate food and healthcare services. These socioeconomic barriers can manifest as insufficient nutrient intake and delayed medical intervention, thereby collectively increasing the risk of malnutrition.

In the multivariate logistic regression, the WBC was not statistically significant, but it ranked first in terms of contribution in the random forest model (Figure 2D). We believe the possible reason for this is that multivariate logistic regression calculates regression coefficients based on linear assumptions, placing more emphasis on interpretability and highlighting linear relationships, whereas random forests assess feature contributions to predictions, prioritizing predictive performance and providing some ability to identify non-linear relationships and interactions. Additionally, the former determines significance, quantifies actual impact, and interprets key factors by incorporating confidence intervals, whereas the latter first confirms whether importance is stable, validates it with domain knowledge, and finally quantifies the actual impact to explain key factors. As a result, the WBC produces different results depending on the algorithm used, but it still has some indicative value.

Numerous studies have studied the relationship between malnutrition in patients with neurological disease and short-term clinical outcomes. Li et al. (7) investigated stroke-associated pneumonia (SAP) using logistic regression models. They demonstrated that malnutrition was common in ischemic stroke patients and was associated with an increased risk of stroke-associated pneumonia, suggesting that early nutritional assessment tools could help identify high-risk SAP patients. Cai et al.'s logistic regression analysis showed that the risk of malnutrition correlated with poor outcomes at three months (15, 21). Our study employed Kruskal-Wallis test and multiple linear regression to analyze impact of malnutrition on the costs and duration of hospitalization. The results showed progressively increasing costs as the severity of malnutrition worsened, probably due to greater use of medical resources for recovery. Longer length of stay in malnourished patients may reflect extended rehabilitation needs. Furthermore, the study by Ruiz et al. (32) showed that malnutrition leaded to increased hospitalization costs, while Correia and Waitzberg identified malnutrition as an independent risk factor for prolonged hospital stays and higher medical costs, findings that are consistent with our results (2).

However, the current study has limitations. We did not assess the representativeness of the sample or use a national sample with fully randomized data. The determination of weight loss was based on patients' self-reported recall at the six-month mark, which introduces subjective bias into the assessment of malnutrition. To maximize the accuracy of this data, we excluded patients with psychiatric or memory disorders; however, this approach may have affected the enrollment of conditions such as Parkinson's disease and Alzheimer's disease. Lack of data on National Institute of Health Stroke Scale (NIHSS) scores, comorbidities, and medication use prevented a complete analysis of the impact of confounders on malnutrition and hospitalization outcomes. Due to the cross-sectional nature of this study, the causal relationship between the influencing factors and malnutrition could not be tested. Future research should assess sample representativeness, reduce subjectivity in weight reporting, compare different assessment methods for national assessment of malnutrition in neurological disease, and conduct prospective studies to further investigate these causal relationships.

In conclusion, our study reveals the malnutrition status and GLIM indicator distribution of neurological disease in China. The results demonstrate the high prevalence of malnutrition in these patients, with neutrophil percentage, lymphocytes, RBC, western region, and education level as influencing factors. Malnutrition may increase the length of stay and hospitalization costs. Given its high incidence and impact on healthcare resources, clinical practice should prioritize assessment of nutritional status on admission. Assessments should take full account of laboratory indicators and individual variability, recognizing differences between methods. The GLIM criteria adopted here provide a more comprehensive assessment of malnutrition.