Traumatic brain injury (TBI) is a leading cause of global mortality and disability. Early identification of mortality risk factors is crucial for improving outcomes. The Braden Scale score, a routinely collected nursing assessment tool originally designed for pressure injury risk assessment, has been underexplored for mortality prediction in TBI patients. This study aimed to evaluate the association between Braden Scale scores at ICU admission and short-term all-cause mortality in TBI patients. This retrospective cohort study utilized the MIMIC-IV database as the derivation cohort and the eICU-CRD database for external validation. Adult TBI patients with complete Braden Scale assessments and outcome data were included. The primary exposure was the initial Braden Scale score, categorized into low and high groups based on the median value. Outcomes included 28 days and 60 days all-cause mortality in both hospital and ICU settings. We also evaluate the components of braden scores by cox regression for outcoms. Statistical analyses included multivariable Cox regression, restricted cubic splines, subgroup analyses, and assessments of predictive performance using IDI, NRI, ROC curves, and decision curve analysis. A total of 2355 patients were analyzed. The low Braden score group showed significantly higher mortality across all endpoints: hospital 28 days (48.1% vs. 19.7%), hospital 60 days (49.4% vs. 20.7%), ICU 28 days (46.2% vs. 18.9%), and ICU 60 days (47.6% vs. 19.8%) (all p < 0.001). After multivariable adjustment, low Braden scores remained independently associated with mortality, with adjusted hazard ratios ranging from 1.78 to 1.84. In a cox regression analysis for the components of Braden Scale, we identified significant inverse associations of sensory perception, activity, nutrition, and friction/shear with multiple mortality outcomes. A nonlinear dose–response relationship was observed, with mortality risk increasing sharply at scores below 12–15. Subgroup analyses indicated a more pronounced association in patients aged ≤ 65 years, while the association was attenuated in those with renal disease. The Braden score demonstrated robust discriminative ability (AUCs: 0.77–0.82) and clinical utility in decision curve analysis. The Braden Scale score is a strong, independent, and readily available predictor and risk factor of short-term mortality in TBI patients, demonstrating a nonlinear inverse relationship with death risk. Subitem analysis further identified sensory perception, activity, nutrition, and friction/shear as key components significantly associated with reduced mortality. These properties support its practical utility for risk stratification at the bedside.

Critically ill patients with spinal cord injury (SCI) and those at high risk for sarcopenia represent vulnerable populations with heightened mortality risks. Traditional prognostic scores often fail to capture the unique pathophysiology of these patients. This study aimed to develop and validate machine learning-based models using easily obtainable clinical features to stratify mortality risk for patients with SCI, high-risk sarcopenia, or both conditions. This retrospective observational study utilized data from the MIMIC-IV (training set; n = 7904) and eICU-CRD (validation set) databases. Patient cohorts were identified based on diagnostic codes for spinal cord injury and high-risk features for sarcopenia, as per established expert consensus. A comprehensive set of clinical variables was extracted, including demographics, comorbidities, ICU severity scores, and laboratory parameters. The primary outcome was all-cause mortality. To construct a prognostic model, we employed an integrated computational approach encompassing ten machine learning algorithms, generating 101 predictive models. The optimal model was selected based on the highest Harrell’s C-index under 10-fold cross-validation. Patients were stratified into high- and low-risk groups using the model-derived risk score. Predictive performance was assessed using the C-index, with survival differences between risk groups validated by Kaplan-Meier analysis. The final model was presented as an interactive nomogram, and its clinical utility was evaluated using calibration curves, decision curve analysis, and time-dependent receiver operating characteristic curves. Statistical analyses included Cox regression and subgroup analyses to ensure the robustness of the findings. A total of 1547 patients with SCI and 4981 patients at high risk of sarcopenia were included. Sarcopenia was significantly associated with increased mortality among SCI patients. Across all cohorts, eight overlapping prognostic features—age, INR, PT, PTT, anion gap, hematocrit, hemoglobin, and platelet count—were identified as key determinants of survival. The RSF model demonstrated robust discrimination in the training cohort (C-index range: 0.88–0.91) and acceptable performance in the external validation cohort (C-index range: 0.62–0.70). Patients stratified into high-risk groups consistently exhibited significantly worse survival outcomes. Each 1-SD increase in the standardized risk score was strongly associated with mortality in both databases. Compared with the Charlson Comorbidity Index, OASIS, and Glasgow Coma Scale, the proposed risk score achieved significantly superior predictive performance, with marked improvements in NRI and IDI. Nomograms derived from Cox regression showed excellent calibration, strong discrimination, and favorable clinical net benefit across all three patient cohorts. We developed and externally validated robust, interpretable machine learning–based prognostic signatures for mortality prediction in patients with SCI, high-risk sarcopenia, and their co-occurrence. The proposed risk score and nomograms outperform traditional clinical scoring systems and provide reliable tools for individualized risk stratification and clinical decision-making in critically ill populations. The online version contains supplementary material available at 10.1186/s12911-026-03408-1.

Dementia presents a significant global health burden, with an increasing prevalence and considerable economic implications. In the intensive care unit (ICU), the number of dementia patients has steadily increased, highlighting the need for more accurate risk stratification. Traditional methods of patient classification often assume homogeneity, limiting their ability to identify subgroups at varying levels of mortality risk. This study explored the potential of graph attention networks (GATs) in addressing this gap, aiming to better characterize high-risk dementia subgroups in the ICU and improve clinical outcomes. This study utilized anonymized patient data from the publicly available MIMIC-IV and MIMIC-III datasets. Dementia patients were identified using ICD-9 and ICD-10 codes, with exclusions based on specific admission criteria, resulting in the inclusion of 7904 patients. A comprehensive set of demographics, comorbidities, ICU severity scores, and laboratory parameters was selected. Three unsupervised clustering methods were employed: a graph autoencoder (GAE), a graph attention autoencoder (GAT), and K-means combined with UMAP-based dimensionality reduction. The optimal number of clusters (k = 3) was determined by maximizing the silhouette scores and Calinski–Harabasz indices. The GAT method quantitatively outperformed both the GAE and K-means methods, achieving superior cluster separation (MIMIC-IV: highest silhouette score: 0.090; Calinski–Harabasz: 345.501; Davies–Bouldin: 2.30; MIMIC-III: highest silhouette score: 0.210; Calinski–Harabasz: 380.620; Davies–Bouldin: 1.850). Three distinct subgroups were identified: Cluster 1 represented the highest-risk phenotype, characterized by advanced age, coagulopathy, anemia, and organ dysfunction, with a strikingly elevated mortality risk (HR = 21.58, p < 0.001). Compared with the unstratified cohort, the machine learning models trained on the MIMIC-IV and validated on the MIMIC-III cohort showed improved predictive performance (AUC up to 0.85 within clusters). SHAP analysis revealed cluster-specific drivers, including a prolonged length of stay, renal impairment, and inflammatory markers. Visualization of the subgroups was achieved via sunburst plots, radar charts, and boxplots. This study included 7904 patients, who were divided into two groups: the nonoccurrence group (n = 5,038) and the occurrence group (n = 2,866). Baseline characteristics, including demographic, clinical, and laboratory factors, significantly differed between the groups. Unsupervised clustering methods identified distinct patient subgroups, with the GAT algorithm outperforming K-means and GCN in clustering both cohorts. Cox regression revealed that the GAT identified high-risk groups, with mortality significantly higher in Cluster 1 (HR = 21.58, P < 0.001). Cluster analysis revealed that Cluster 1 had the highest mortality risk and was characterized by advanced age, severe renal dysfunction, and poor clinical outcomes. To validate the superiority of the GAT-based unsupervised clustering approach, we applied various machine learning algorithms, which demonstrated enhanced predictive performance across the identified subgroups. The feature distributions and interrelationships were visualized using radar plots, boxplots, and SHAP analysis. SHAP further revealed distinct risk profiles: Cluster 0 was associated with an increased risk of thrombotic events and acute kidney injury; Cluster 1 reflected a high metabolic stress state indicative of early systemic inflammatory response syndrome (SIRS) or sepsis; and Cluster 2 was primarily marked by severe renal failure. This study demonstrates the effectiveness of GAT-based clustering for identifying high-risk subgroups among dementia patients in the ICU. By providing more granular risk stratification, these models can support clinical decision-making and improve patient management. These findings underscore the potential of advanced machine learning techniques in enhancing ICU care for dementia patients, with future research focusing on temporal validation and integration into clinical workflows to optimize outcomes.