Machine learning–based risk stratification for early tracheotomy in geriatric intracerebral hemorrhage: model development and web deployment

Machine learning–based risk stratification for early tracheotomy in geriatric intracerebral hemorrhage: model development and web deployment | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Machine learning–based risk stratification for early tracheotomy in geriatric intracerebral hemorrhage: model development and web deployment Jiayin Wang, Zhenyu Fan, Longjie Chen, Shanglu Lin, Gang Wang, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8177019/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Respiratory complications are common among elderly patients with intracerebral hemorrhage (ICH), which can cause more hypoxemia and neuronal injury, further leading to poor prognosis. Tracheotomy (TT) can effectively improve airway protection and oxygenation, but the timing of tracheotomy is usually determined by the experience of the attending clinician and lacks objective and standardized decision criteria. A reliable prediction tool for the early identification of elderly ICH patients who need TT is urgently needed. This study aimed to establish and validate a machine learning (ML) model to predict the early application of tracheotomy in elderly ICH patients. Methods A total of 435 elderly patients (≥ 65 years) with ICH from two tertiary hospitals between 2015 and 2023 were retrospectively reviewed. A total of 40 clinical, laboratory, and imaging features were collected at admission. We applied LASSO regression for feature selection. Six machine learning algorithms (Random Forest, XGBoost, SVM, Gradient Boosting, Logistic Regression, and KNN) were trained and tested with a train–test split of 70/30. The imbalanced class distribution was handled by SMOTE and class-weight adjustments. Hyperparameter optimization was conducted with Bayesian optimization. The performance of each algorithm was evaluated based on ROC–AUC, PR curves, accuracy, precision, recall, and F1 score. SHAP values were calculated to interpret the models. The best-performing model was deployed as an online prediction tool. Results LASSO identified five predictors: Glasgow Coma Scale (GCS) score, intraventricular hemorrhage (IVH), midline shift, albumin level, and CT biphasic sign. Among the six algorithms, Random Forest performed the best (composite score 0.831; test AUC 0.723; accuracy 0.793; precision 0.784; recall 0.793; F1 = 0.788). SHAP analysis identified that GCS was the most important feature that contributed to tracheotomy risk, followed by IVH and midline shift. An online prediction tool was successfully deployed to estimate the early risk of tracheotomy in real time. Conclusions A ML–based model with five readily available clinical features accurately predicted the early need for TT in elderly ICH patients and may be helpful in timely airway management decisions. Intracerebral hemorrhage Tracheotomy Elderly Machine learning Random Forest SHAP Prediction model Airway management Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Intracerebral hemorrhage (ICH) is one of the most severe causes of stroke and is associated with the highest mortality and disability rates among the elderly. In older patients, less physiological reserve, decreased compensation ability, and multiple comorbidities may result in worse neurological outcomes than in younger patients[1, 2]. Aspiration pneumonia, airway obstruction, and hypoventilation, respiratory complications occur more than 60% of ICH patients and are an important cause of secondary hypoxemia[3]. Hypoxemia can aggravate brain tissue hypoxia, exacerbate perihematomal edema, and lead to adverse changes in intracranial pressure (ICP), ultimately resulting in secondary brain injury[4]. The above complications are more common in the elderly. Weakened cough reflex, airway protection ability, and lung compliance, leading to a high incidence of respiratory failure[5]. Timely and appropriate management of airway protection is critical to prevent secondary brain injury and improve outcomes of elderly ICH patients. Tracheotomy (TT) is a common surgical procedure to provide airway protection, improve ventilation, prevent aspiration, and reduce hypoxemia in patients with severe neurological deficits[6]. TT can shorten sedation time, facilitate pulmonary toileting, and reduce the incidence of ventilator-associated pneumonia and other complications when performed early[7]. However, the optimal timing for tracheotomy has not been well established and may vary depending on the attending physician’s experience and clinical judgment. There is no widely accepted consensus or standardized timing window for TT, leading to delayed treatment, respiratory deterioration, and secondary neurological injury[8, 9]. Although several clinical predictors, such as Glasgow Coma Scale (GCS) score, hematoma volume, and brainstem compression, have been proposed to guide clinical decision-making, there is no validated tool to help clinicians identify ICH patients who would benefit from TT the most. Studies specifically evaluating these predictors in the elderly ICH population are even more serious. Although several previous studies have evaluated the risk factors of TT and tried to determine the optimal timing[10–12], there were some important limitations that should be taken into consideration. The majority of studies were from single centers and had relatively small samples without external validation or more diverse patient cohorts[11]. Some studies used traditional linear or logistic regression models, which were suboptimal for capturing complex nonlinear relationships between features and the target. More importantly, most of the studies did not focus on older ICH patients, who are more vulnerable to airway complications but also have different clinical characteristics. To overcome these limitations, we performed a multicenter analysis of elderly ICH patients and utilized several machine learning (ML) methods including LASSO feature selection, Random Forest (RF) classification, Bayesian optimization, and SHAP interpretability. Compared with previous studies, this design can lead to more accurate prediction, better robustness, and improved interpretability for clinical use. We also developed and implemented a web-based tool for clinical use based on the prediction model, which could serve as an evidence-based clinical decision support system to help clinicians identify ICH patients with high risk for tracheotomy, particularly for the elderly. This study was a multicenter, retrospective cross-sectional investigation reported in accordance with the STROBE guidelines. It aimed to explore the association between clinical characteristics, relevant indicators, and outcomes—specifically whether elderly patients with ICH underwent tracheotomy. Methods and Materials Study Population The study population was drawn from two tertiary teaching hospitals in China. Specifically, it included patients admitted to the Second Hospital of Lanzhou University between March 2022 and December 2023, and patients admitted to the Second Affiliated Hospital of Fujian Medical University between January 2015 and April 2022 who were diagnosed with ICH. A total of 1,394 patients were initially screened, among whom 435 elderly patients aged ≥65 years met the eligibility criteria and were included as the final analytical cohort. All enrolled patients were diagnosed with spontaneous ICH based on the International Classification of Diseases, Ninth Revision (ICD-9) or Tenth Revision (ICD-10), according to standard diagnostic criteria. The exclusion criteria were as follows: 1) Age <65 years; 2) Secondary ICH due to trauma, brain tumors, aneurysms, or arteriovenous malformations; 3) Patients with multiple hospitalizations for ICH, for whom only the first admission was considered; 4) Severe hepatic or renal dysfunction, leukemia, lymphoma, other hematologic disorders, or malignancies; and 5) Incomplete laboratory data. Ultimately, 435 elderly patients who met all inclusion criteria were enrolled in the study (Figure 1). Ethical Statement Ethical approval was obtained from the Ethics Committee of the Second Hospital & Clinical, Medical School, Lanzhou University (Approval No. 2025A-1073) and the Ethics Committee of the Second Affiliated Hospital of Fujian Medical University (Approval No. 2024-016). The study strictly adhered to the ethical principles of the Declaration of Helsinki. All research data were anonymized and de-identified before use, and the requirement for written informed consent was waived by the ethics committees. Variable Collection A comprehensive set of predictor variables associated with the outcome of TT was collected based on prior literature, including demographic characteristics, lifestyle factors, comorbidities, clinical symptoms, neuroimaging features, and laboratory indicators. Demographic variables consisted of age and sex; lifestyle factors included current smoking and alcohol consumption. Comorbidities relevant to intracerebral hemorrhage, such as hypertension and diabetes mellitus, were also recorded. Clinical status variables included GCS score, body temperature, respiratory rate, heart rate, systolic and diastolic blood pressure. Neuroimaging variables included the location of hemorrhage (lobar, basal ganglia, thalamus, cerebellum, or brainstem), presence of intraventricular hemorrhage (IVH), CT hematoma signs (none, island sign, blend sign, fluid level sign, or black hole sign), laterality of the hematoma, midline shift, and initial hematoma volume. Hematoma volume was calculated by radiologists using manual segmentation in 3D Slicer software. Laboratory data were obtained from the patients’ initial admission tests. These included glucose, uric acid, electrolytes (potassium, sodium, calcium, phosphorus, magnesium), total protein, albumin, complete blood count parameters (white blood cell count, neutrophils, lymphocytes, monocytes, hemoglobin, platelets), and coagulation indices (prothrombin time [PT], international normalized ratio [INR], activated partial thromboplastin time [APTT], fibrinogen, thrombin time [TT], and D-dimer). Definition of Outcome The primary outcome of this study was early TT, defined as the performance of a TT within 7 days after hospital admission[13, 14]. Patients who did not undergo TT or received the procedure after day 7 were classified into the non-TT group. Data Processing and ML Modeling All patient samples were randomly divided into a training set and a testing set in a 7:3 ratio. Six supervised machine learning algorithms were employed for model development, including logistic regression (LR), RF, extreme gradient boosting (XGBoost), support vector machine (SVM), K-nearest neighbors (KNN), and an ensemble model. These algorithms were selected to balance linear and nonlinear modeling capabilities and to compare performance in terms of structural complexity, interpretability, and computational efficiency. During preprocessing, missing values of continuous variables were imputed using the median, while categorical variables were transformed into numerical form through one-hot encoding to ensure compatibility with all models. All variables were subsequently standardized to minimize potential bias due to differences in measurement scales. Preliminary exploratory analysis detected no abnormal outliers, and the distribution and scale of features were generally consistent, supporting their suitability for model training. Feature selection was performed using the least absolute shrinkage and selection operator (LASSO) regression, with ten-fold cross-validation used to determine the optimal regularization parameter λ. A coefficient trace plot was generated to visually demonstrate the feature shrinkage process. Considering the imbalance in the original dataset (with relatively few patients receiving tracheotomy), we applied the Synthetic Minority Oversampling Technique (SMOTE) together with class weight adjustments during model training to mitigate bias toward the majority class. Hyperparameters for each algorithm were optimized using Bayesian optimization embedded within five-fold cross-validation, which enhances model stability and generalization while reducing redundant parameter search and computational cost. The final optimal hyperparameters were: max_depth = 17, max_features = ‘log2’, min_samples_leaf = 1, min_samples_split = 2, and n_estimators = 107. Comprehensive model performance was evaluated using receiver operating characteristic (ROC) curves and area under the curve (AUC), precision–recall (PR) curves, calibration curves, as well as accuracy, precision, recall, and F1 score. Model Interpretation and Web Deployment To address the “black-box” limitation commonly associated with traditional machine learning models, we incorporated Shapley Additive Explanations (SHAP) to quantify the contribution of each feature to the prediction and to enhance interpretability. SHAP provides both global and individual-level explanations, allowing us not only to understand overall feature importance trends but also to interpret predictions for specific patients in a personalized manner. To further improve clinical usability and operational convenience, we deployed the optimized predictive model as an online application using the Flask framework. Clinicians can obtain a real-time prediction of a patient's tracheotomy risk by simply entering the required key feature values. Importantly, the prediction platform has been made freely accessible to both healthcare professionals and patients, thereby enhancing the translational potential and real-world applicability of the model. Statistical Analysis Categorical variables were compared between groups using the chi-square test or Fisher’s exact test, as appropriate. Continuous variables were analyzed using the independent samples t-test or the Mann–Whitney U test based on the results of normality testing. Model performance was assessed using multiple evaluation metrics, including the AUC, accuracy, precision, recall, and F1 score. The AUC values of the training and testing sets were used as the primary indicators of predictive performance. A two-sided p-value ≤ 0.05 was considered statistically significant. All data processing and statistical analyses were conducted within a conda virtual environment in PyCharm. Results Baseline Characteristics A total of 435 elderly patients with ICH were included in this study, of whom 83 underwent TT (TT group) and 352 did not (Non-TT group), as shown in Table 1. Baseline comparisons revealed significant differences between the two groups across multiple variables, which may influence or reflect the risk of requiring TT. Patients in the TT group exhibited more severe neurological impairment, including lower GCS scores, larger initial hematoma volumes, and more pronounced midline shift. In addition, inflammatory markers were elevated in the TT group, with higher white blood cell and neutrophil counts, whereas lymphocyte counts were lower compared with the Non-TT group. Blood glucose levels were also significantly higher in the TT group. Regarding underlying comorbidities, hypertension was more prevalent in the TT group. The biphasic CT sign appeared more frequently as well. Moreover, basal ganglia hemorrhage and IVH were markedly more common in the TT group, indicating that these patients tended to have more complex and severe disease, which may predispose them to requiring TT. The dataset was divided into training and testing sets at a 7/3 ratio, and comparison showed good balance between the two subsets (Supplementary Table 1). Feature Selection and Variable Contribution To initiate the modeling process, feature selection was performed using all patient data within the training set. Key predictors were identified through the least absolute shrinkage and selection operator (LASSO) regression, with ten-fold cross-validation used to determine the optimal regularization parameter α. The LASSO coefficient path plot (Figure 2A) illustrates the dynamic shrinkage of feature coefficients as α changes. The optimal α value of 0.0118 (Figure 2B) was selected based on the minimum mean squared error (MSE), yielding a stable and high-performing model while effectively reducing multicollinearity. A total of five significant predictors were retained after LASSO screening: GCS score, IVH, midline shift, serum albumin level, and the CT blend sign (Figure 2C). These features were subsequently used for model development. Correlation analysis (Figure 2D) showed that all pairwise correlation coefficients were below 0.6, suggesting no considerable multicollinearity among the selected variables. Residual distribution analysis (Figure 2E) demonstrated an approximately normal distribution, supporting the good fit and statistical stability of the LASSO model. Furthermore, density distribution plots indicated that the five key features showed consistent patterns in value distribution and central tendency, providing a solid foundation for exploring their relationships with disease phenotypes and clinical outcomes. ML Model Testing To predict the risk of tracheotomy among elderly patients with ICH, we evaluated six supervised machine learning algorithms: RF, XGBoost, Extra Trees, SVM, Gradient Boosting, and Logistic Regression. The performance of each model was assessed across multiple metrics, including AUC, accuracy, precision, recall, and F1 score. Visualization was conducted using comprehensive scoring plots, radar charts, ROC curves, PR curves, and bar charts. Among all models, the RF algorithm demonstrated the best overall performance. In terms of composite scores, the RF model achieved the highest value (0.831), significantly outperforming the other algorithms (Figure 3A). The ROC curve showed that Random Forest had the largest AUC (0.687), surpassing XGBoost (0.647) and SVM (0.658) (Figure 3C). The PR curve further revealed that Random Forest maintained a favorable balance between precision and recall compared with other models (Figure 3D). Radar charts (Figure 3B) and bar plots (Figure 3E) consistently confirmed that the Random Forest model ranked highest in accuracy, recall, and F1 score. In summary, these findings indicate that RF provided the most robust and reliable predictive performance among the six tested algorithms and was therefore selected as the core model for subsequent optimization and deployment. Model Optimization and Evaluation To further enhance predictive performance and generalizability, Bayesian optimization was employed to automatically tune the key hyperparameters of the model. Using ten-fold cross-validation with the average ROC–AUC as the objective function, the algorithm efficiently identified the optimal parameter combination: max_depth = 17, max_features = “log2”, min_samples_leaf = 1, min_samples_split = 2, and n_estimators = 107. After determining the optimal parameters, the final RF model was retrained on the full training set and subsequently evaluated on the independent testing set. The ROC curve demonstrated improved predictive accuracy after tuning, with an AUC of 0.723 (Figure 4A). The precision–recall curve showed a PR-AUC of 0.447, and precision gradually decreased as recall increased, indicating that the model preserved good balance in minority-class prediction, which is crucial for identifying high-risk individuals (Figure 4B). The calibration curve revealed strong agreement between predicted probabilities and observed outcomes, closely approximating the ideal diagonal line and suggesting reliable probability estimation (Figure 4C). Additional performance metrics showed that the optimized Random Forest model achieved balanced results across all indices (Figure 4D). The heatmap (Figure 4E) summarized the key performance metrics: accuracy (0.793), precision (0.784), recall (0.793), F1 score (0.788), and AUC (0.723), collectively demonstrating the model’s stability, robustness, and potential for clinical application in predicting tracheotomy risk. SHAP-Based Model Interpretability Analysis To enhance the clinical interpretability of the predictive model, we applied SHapley Additive exPlanations (SHAP) to systematically analyze the contribution of each key feature to tracheotomy risk. The global SHAP beeswarm plot (Figure 5A) demonstrated that the GCS score was the most influential factor, with lower GCS values associated with markedly increased SHAP values, indicating a higher likelihood of tracheotomy. IVH and midline shift also contributed positively to risk prediction, reflecting the importance of structural brain injury. Lower serum albumin levels were associated with increased risk, whereas the CT blend sign had a relatively smaller but still notable effect. The SHAP feature importance ranking (Figure 5B) further confirmed the overall contributions of GCS, IVH, midline shift, albumin, and CT hematoma signs in descending order of importance. The SHAP decision plot (Figure 5C) illustrated how each feature sequentially influenced the predicted risk, consistent with the patterns observed in global explanations. Additionally, SHAP waterfall plots were generated for two representative cases: one patient who underwent tracheotomy (Figure 5D) and one who did not (Figure 5E). These plots clearly showed the positive (risk-increasing) and negative (risk-reducing) contributions of individual features to each patient’s prediction. In the figures, f(x) represents the model’s predicted output, and E[f(x)] represents the baseline expected value. Red bars denote risk-enhancing features, while blue bars represent protective factors, with bar length indicating the magnitude of influence. Overall, the SHAP analysis substantially improved the transparency of the model’s decision-making process and supports personalized clinical risk assessment and decision-making. Online Web Deployment Figure 6 showed that to enhance the clinical applicability of the model, we developed and deployed an online tracheotomy risk prediction system based on the optimized RF algorithm using the Flask framework. The platform (http://47.109.146.0:8000/) allows users to obtain real-time predictions by simply entering five core clinical features. The system automatically generates the predicted probability of early TT for patients with ICH, thereby improving accessibility and convenience for clinical decision-making. As an example, for a representative patient with a GCS score of 15, serum albumin level of 34.3 g/L, midline shift of 4 mm, positive CT blend sign, and presence of intraventricular hemorrhage, the system estimated a tracheotomy probability of 76.64%, with a 95% confidence interval of [68.62%, 84.65%]. The predicted probability and confidence interval are displayed immediately on the right side of the interface, offering an intuitive and user-friendly tool for clinicians. Discussion This multicenter cohort study among older adults with ICH developed and validated a ML–based clinical prediction model for the support of early identification of ICH patients who would most likely benefit from early TT. Leveraging the integration of routinely available clinical, imaging, and laboratory variables, this model can offer more accurate and individualized risk stratification than experience-based clinical decision making. The practical clinical utility is that the web-based ML model can estimate the individualized TT risk at admission, and thus, if implemented, help clinicians in optimizing airway management in this patient population, avoiding secondary hypoxemia, and improving the outcomes among these patients. Previous models of predictors for TT or prolonged ventilation in neurocritical populations have several limitations. While earlier studies have established factors such as low GCS scores, intraventricular hemorrhage, and large hematoma burden as being more likely to undergo tracheotomy, these were mostly developed using cohorts with small sample sizes from single centers as these research by Shadi Yaghi et al. and Szeder et al.[10, 11]. Later models, including the commonly used SET score which was built by Silvia et al.[15], attempted to incorporate neurological, radiological, and systemic variables and demonstrated moderate discrimination in stroke and ICH populations[12, 15]. However, external validations of these scores have consistently shown moderate discrimination at best, with variation between centers and patient subgroups. Older adults, a key focus of this study, are underrepresented in most previous prediction models of prolonged mechanical ventilation after ICH[16]. In comparison with previous efforts, our study has several novel characteristics. First, we focused specifically on the ICH population in geriatric patients, a group which is at particularly high risk of developing respiratory complications after ICH yet to date lacks a dedicated prediction tool. Second, by using a ML approach rather than a predefined scoring system, we were able to capture nonlinear interactions between variables. Third, we provided feature explainability, and released the model as an online calculator to better support its clinical applicability in guiding more individualized airway-management decisions in older adults. Predictors included in the RF model are also biologically plausible. Low GCS score suggests decreased level of consciousness and airway protective reflexes and can predispose to aspiration, hypoventilation, and hypoxemia[17]. IVH and midline shift indicate severe structural brain injury that may lead to a deeper sedation level and longer duration of mechanical ventilation[18, 19]. Hypoalbuminemia could suggest the presence of systemic inflammatory response and malnutrition or poor nutritional reserves in critically ill patients as well as a state of increased catabolism, which is a condition of high metabolic demand in the human body[20, 21]. Therefore, it can be a predisposing factor for poor weaning outcome. Biphasic CT sign was associated with hematoma heterogeneity and early hematoma rapid expansion, which could lead to indirect neurological compromise[22]. It is of note that most of these clinical and radiographic predictors of post-ICH respiratory deterioration were statistically significant and biologically plausible, as included in our model. ML has the potential to combine a number of multidimensional predictors of outcomes in an interpretable way that aligns with the actual thought process of clinical care in the real world more so than looking at individual parameters. In a geriatric medicine view, this model may help identify older adults at higher risk of post-ICH respiratory deterioration earlier than clinical judgment alone and thus will be able to determine whether the need for a TT is likely, which will hopefully allow earlier intervention and decreased secondary hypoxic injury and in-hospital death as well as higher functional outcome. There are several clinically relevant takeaways from this work. First, we provide an evidence-based tool to inform clinicians about the risk of early TT for a given older patient. While physicians commonly make airway management decisions, those decisions and the timing of them can be highly subjective and variable between clinicians. A risk prediction model to aid in airway decision-making can help address such limitations and promote standardization, equity, and timeliness. This study hypothesize that knowing ahead of time the risk of early tracheotomy may also allow for earlier anticipation, family discussion, allocation of neurocritical care resources, and earlier respiratory care interventions that might limit secondary hypoxemia and preventable complications. Second, the risk prediction model produces individualized risk estimates, at the time of admission, that can aid in airway management decisions. These patient-level predictions can be instrumental in the growing field of precision geriatric medicine, in which intervention selection is individualized based on physiologic rather than chronological age. Third, this approach provides a user-friendly web-based application, which can be accessed at the bedside with no need for specialty software, that may promote real-world uptake and use in clinical care. This study has several limitations that need to be addressed. First, the data were collected from two large tertiary referral hospitals, and the retrospective design may limit the generalizability of the results and introduce selection bias. Clinical decision-making patterns may vary across institutions, and external validation in broader and more community-based geriatric populations will be necessary to ensure the robustness of the model. Second, while the model included important clinical, imaging, and laboratory variables, other potential predictors, such as frailty status, pre-morbid functional status, cognitive impairment, and detailed cardiopulmonary assessments, were not consistently available and could be further explored to enhance the predictive performance. Third, the ML methods still rely on the quality and completeness of the data, and their results may be affected by unmeasured confounders. Prospective studies with standardized data collection, predefined TT criteria, and a focus on assessing the clinical utility of the model are needed. Future research could also consider integrating longitudinal physiological data, continuous monitoring signals, and frailty-informed geriatric assessments. Finally, implementation studies are necessary to understand how the prediction tool can be integrated into clinical workflows, impact decision-making timing, and improve patient-centered outcomes. Conclusion In this multicenter cohort of older adults with intracerebral hemorrhage, the research developed and validated an interpretable ML model that predicted early TT using five easily obtained admission variables. The model had balanced performance and enhanced predictive accuracy compared with traditional clinical methods. By combining important neurological, imaging, and laboratory markers, and providing an easily accessible web-based calculator, this study offers a practical decision-support tool that may help clinicians identify high-risk patients earlier and refine airway management strategies. These results underscore the potential of data-driven approaches to facilitate personalized care and reduce secondary respiratory complications in geriatric neurocritical populations. Declarations Acknowledgement We thank the clinical teams at Lanzhou University Second Hospital and Fujian Medical University Second Affiliated Hospital for their support in data retrieval and verification. We also acknowledge the contributions of the patients and their families who made this research possible. Author Contributions WJY (First Author), CLJ: (Co-first Author) Conceptualization, methodology, data curation, formal analysis, writing—original draft. FZY (Co-first Author): Data collection, imaging review, statistical validation. LSL: Machine-learning modeling, software development, visualization. WG: Clinical interpretation, manuscript review, supervision. HJZ (Co-corresponding Author): Project administration, resource coordination, writing—review & editing. QWZ (Corresponding Author): Study design, critical revision, final approval, and correspondence. Ethics Statement This study was approved by the Ethics Committee of the Second Hospital & Clinical, Medical School, Lanzhou University (Approval No. 2025A-1073) and Second Affiliated Hospital of Fujian Medical University (Approval No. 2024-016). All methods adhered to the Declaration of Helsinki. Because this was a retrospective analysis of anonymized data, the requirement for written informed consent was waived. Human Ethics and Consent to Participate Declarations Not applicable. Clinical Trial Number Clinical trial number: not applicable. Competing Interests Statement The authors declare that they have no competing interests. Data Availability Statement The datasets generated and analyzed during the current study are not publicly available due to institutional restrictions but are available from the corresponding author on reasonable request. Funding Statement Joint funds for the innovation of science and technology, Fujian province(Grant number:2023Y9239） Guiding Project of Quanzhou Science and Technology Bureau (Grant number: 2023N064S) References Ribo M, Grotta JC. Latest advances in intracerebral hemorrhage. Curr Neurol Neurosci Rep. 2006;6:17–22. https://doi.org/10.1007/s11910-996-0004-0. Gaist D, Wallander M-A, González-Pérez A, García-Rodríguez LA. Incidence of hemorrhagic stroke in the general population: validation of data from the health improvement network. Pharmacoepidemiol Drug Saf. 2013;22:176–82. https://doi.org/10.1002/pds.3391. Maramattom BV, Weigand S, Reinalda M, Wijdicks EFM, Manno EM. Pulmonary complications after intracerebral hemorrhage. Neurocrit Care. 2006;5:115–9. https://doi.org/10.1385/NCC:5:2:115. Yang Y, Rosenberg GA. Blood-brain barrier breakdown in acute and chronic cerebrovascular disease. Stroke. 2011;42:3323–8. https://doi.org/10.1161/STROKEAHA.110.608257. Lu P, Cao Z, Gu H, Li Z, Wang Y, Cui L, et al. Association of sex and age with in-hospital mortality and complications of patients with intracerebral hemorrhage: a study from the chinese stroke center alliance. Brain Behav. 2023;13:e2846. https://doi.org/10.1002/brb3.2846. Bösel J, Niesen W-D, Salih F, Morris NA, Ragland JT, Gough B, et al. Effect of early vs standard approach to tracheostomy on functional outcome at 6 months among patients with severe stroke receiving mechanical ventilation: the SETPOINT2 randomized clinical trial. JAMA. 2022;327:1899–909. https://doi.org/10.1001/jama.2022.4798. Ding W-L, Xiang Y-S, Liao J-C, Wang S-Y, Wang X-Y. Early tracheostomy is associated with better prognosis in patients with brainstem hemorrhage. J Integr Neurosci. 2020;19:437–42. https://doi.org/10.31083/j.jin.2020.03.25. Ko S-B, Choi HA, Parikh G, Helbok R, Schmidt JM, Lee K, et al. Multimodality monitoring for cerebral perfusion pressure optimization in comatose patients with intracerebral hemorrhage. Stroke. 2011;42:3087–92. https://doi.org/10.1161/STROKEAHA.111.623165. Imperiale C, Magni G, Favaro R, Rosa G. Intracranial pressure monitoring during percutaneous tracheostomy “percutwist” in critically ill neurosurgery patients. Anesth Analg. 2009;108:588–92. https://doi.org/10.1213/ane.0b013e31818f601b. Szeder V, Ortega-Gutierrez S, Ziai W, Torbey MT. The TRACH score: clinical and radiological predictors of tracheostomy in supratentorial spontaneous intracerebral hemorrhage. Neurocrit Care. 2010;13:40–6. https://doi.org/10.1007/s12028-010-9346-1. Yaghi S, Moore P, Ray B, Keyrouz SG. Predictors of tracheostomy in patients with spontaneous intracerebral hemorrhage. Clin Neurol Neurosurg. 2013;115:695–8. https://doi.org/10.1016/j.clineuro.2012.08.010. Alsherbini K, Goyal N, Metter EJ, Pandhi A, Tsivgoulis G, Huffstatler T, et al. Predictors for tracheostomy with external validation of the stroke-related early tracheostomy score (SETscore). Neurocrit Care. 2019;30:185–92. https://doi.org/10.1007/s12028-018-0596-7. Bösel J, Schiller P, Hook Y, Andes M, Neumann J-O, Poli S, et al. Stroke-related early tracheostomy versus prolonged orotracheal intubation in neurocritical care trial (SETPOINT): a randomized pilot trial. Stroke. 2013;44:21–8. https://doi.org/10.1161/STROKEAHA.112.669895. Robba C, Galimberti S, Graziano F, Wiegers EJA, Lingsma HF, Iaquaniello C, et al. Tracheostomy practice and timing in traumatic brain-injured patients: a CENTER-TBI study. Intensive Care Med. 2020;46:983–94. https://doi.org/10.1007/s00134-020-05935-5. Schönenberger S, Al-Suwaidan F, Kieser M, Uhlmann L, Bösel J. The SETscore to predict tracheostomy need in cerebrovascular neurocritical care patients. Neurocrit Care. 2016;25:94–104. https://doi.org/10.1007/s12028-015-0235-5. Chen X-H, Zhao J-J, Chen C, Yao L. Establishment and validation of a predictive model for tracheotomy in critically ill patients and analysis of the impact of different tracheotomy timing on patient prognosis. BMC Anesthesiol. 2024;24:175–87. https://doi.org/10.1186/s12871-024-02558-x. Parry-Jones AR, Abid KA, Di Napoli M, Smith CJ, Vail A, Patel HC, et al. Accuracy and clinical usefulness of intracerebral hemorrhage grading scores: a direct comparison in a UK population. Stroke. 2013;44:1840–5. https://doi.org/10.1161/STROKEAHA.113.001009. Dai J, Li S, Li X, Xiong W, Qiu Y. The mechanism of pathological changes of intraventricular hemorrhage in dogs. Neurol India. 2009;57:567–77. https://doi.org/10.4103/0028-3886.57798. Jacobs B, Beems T, van der Vliet TM, Diaz-Arrastia RR, Borm GF, Vos PE. Computed tomography and outcome in moderate and severe traumatic brain injury: hematoma volume and midline shift revisited. J Neurotrauma. 2011;28:203–15. https://doi.org/10.1089/neu.2010.1558. Di Napoli M, Behrouz R, Topel CH, Misra V, Pomero F, Giraudo A, et al. Hypoalbuminemia, systemic inflammatory response syndrome, and functional outcome in intracerebral hemorrhage. J Crit Care. 2017;41:247–53. https://doi.org/10.1016/j.jcrc.2017.06.002. Duskin J, Yechoor N, Singh S, Mora S, Senff J, Kourkoulis C, et al. Nutrition markers and discharge outcome in deep and lobar intracerebral hemorrhage. Eur Stroke J. 2024;9:1074–82. https://doi.org/10.1177/23969873241253048. Morotti A, Boulouis G, Dowlatshahi D, Li Q, Shamy M, Al-Shahi Salman R, et al. Intracerebral haemorrhage expansion: definitions, predictors, and prevention. Lancet, Neurol. 2023;22:159–71. https://doi.org/10.1016/S1474-4422(22)00338-6. Table 1 Table 1 The Study Population Demography Tracheotomy Non-TT (N=352) TT (N=83) P-value Age (year) 71.61 ± 5.35 70.72 ± 5.03 0.162 Temperature (℃ ） 36.78 ± 0.51 36.36 ± 3.68 0.749 Respiratory rate (times/minute) 18.94 ± 4.76 18.10 ± 6.03 0.219 Heart rate (times/minute) 83.67 ± 15.95 83.73 ± 17.00 0.53 Systolic blood pressure (mmHg) 160.49 ± 24.08 168.22 ± 28.26 0.004 Diastolic blood pressure (mmHg) 88.29 ± 15.14 91.18 ± 16.98 0.069 Glucose (mmol/L) 7.86 ± 2.87 8.24 ± 2.46 0.04 Uric acid (umol/L) 280.21 ± 120.09 283.82 ± 105.99 0.718 Potassium (mmol/L) 3.67 ± 0.50 3.64 ± 0.47 0.643 Sodium (mmol/L) 138.39 ± 4.08 139.03 ± 4.50 0.156 Calcium (mmol/L) 2.28 ± 0.18 2.24 ± 0.18 0.104 Phosphorus (mmol/L) 0.91 ± 0.31 0.91 ± 0.30 0.629 Magnesium (mmol/L) 0.86 ± 0.13 0.86 ± 0.11 0.449 Total protein (g/L) 70.16 ± 7.98 69.79 ± 9.05 0.595 Albumin (g/L) 38.95 ± 4.30 40.09 ± 5.09 0.025 White blood cell (10 9 /L) 9.89 ± 4.80 10.93 ± 4.16 0.01 Neutrophil (10 9 /L) 8.07 ± 4.63 9.37 ± 4.20 0.002 Lymphocyte (10 9 /L) 1.19 ± 0.68 1.14 ± 0.90 0.047 Monocyte count (10 9 /L) 0.51 ± 0.31 0.51 ± 0.28 0.772 Hemoglobin (g/L) 141.87 ± 21.73 140.76 ± 25.90 0.479 Platelet (10 9 /L) 189.68 ± 88.58 196.73 ± 76.29 0.337 Prothrombin time (s) 12.65 ± 6.67 11.63 ± 0.95 0.527 International normalized ratio 1.06 ± 0.35 0.99 ± 0.09 0.074 Activated partial thromboplastin time (s) 26.33 ± 6.45 26.50 ± 6.38 0.725 Fibrinogen (g/L) 3.20 ± 2.71 3.15 ± 1.04 0.346 Thrombin time (s) 16.79 ± 2.24 16.87 ± 2.04 0.832 dimer (mg/L) 2.61 ± 11.00 1.78 ± 3.20 0.017 Midline shift (mm) 1.85 ± 3.36 4.34 ± 4.39 <0.001 Initial hematoma volume (mL) 23.80 ± 22.45 36.89 ± 20.24 <0.001 GCS 11.06 ± 3.34 8.16 ± 2.83 <0.001 Gender, (%) 0.548 Female 157 (44.60%) 34 (40.96%) Male 195 (55.40%) 49 (59.04%) Hypertension, (%) 0.025 No 102 (28.98%) 14 (16.87%) Yes 250 (71.02%) 69 (83.13%) Diabetes mellitus, (%) 0.056 No 298 (84.66%) 63 (75.90%) Yes 54 (15.34%) 20 (24.10%) Smoking, (%) 0.13 No 296 (84.09%) 64 (77.11%) Yes 56 (15.91%) 19 (22.89%) Alcohol consumption, (%) 0.159 No 316 (89.77%) 70 (84.34%) Yes 36 (10.23%) 13 (15.66%) Anticoagulants, (%) 0.297 No 318 (90.34%) 78 (93.98%) Yes 34 (9.66%) 5 (6.02%) Intraventricular hemorrhage, (%) <0.001 No 242 (68.75%) 34 (40.96%) Unilateral 89 (25.28%) 40 (48.19%) Bilateral 21 (5.97%) 9 (10.84%) Side of hematoma, (%) 0.474 Left 167 (47.44%) 43 (51.81%) Right 185 (52.56%) 40 (48.19%) Lesion location, (%) 0.002 Basal Ganglia 162 (46.02%) 58 (69.88%) Lobar 79 (22.44%) 6 (7.23%) Cerebellum 33 (9.38%) 6 (7.23%) Brainstem 6 (1.70%) 2 (2.41%) Other 26 (7.39%) 2 (2.41%) Thalamus 46 (13.07%) 9 (10.84%) CT signs, (%) <0.001 No 269 (76.42%) 43 (51.81%) Island Sign 23 (6.53%) 9 (10.84%) Biphasic Sign 54 (15.34%) 31 (37.35%) Fluid Leve 4 (1.14%) 0 (0.00%) Black Hole Sign 2 (0.57%) 0 (0.00%) TT: Tracheotomy, GCS: Glasgow Coma Scale, CT: Computed Tomography Additional Declarations No competing interests reported. Supplementary Files suppplementtable1.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8177019","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":565352660,"identity":"78227d8c-382f-4366-943f-06b41df08846","order_by":0,"name":"Jiayin Wang","email":"","orcid":"","institution":"Second Affiliated Hospital of Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Jiayin","middleName":"","lastName":"Wang","suffix":""},{"id":565352661,"identity":"56ddff8a-761a-48d2-b380-5bcc722e5618","order_by":1,"name":"Zhenyu Fan","email":"","orcid":"","institution":"School of Software, North University of China Taiyuan","correspondingAuthor":false,"prefix":"","firstName":"Zhenyu","middleName":"","lastName":"Fan","suffix":""},{"id":565352663,"identity":"a9c0c3cb-b235-4a58-8790-fc26639e3721","order_by":2,"name":"Longjie Chen","email":"","orcid":"","institution":"Second Affiliated Hospital of Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Longjie","middleName":"","lastName":"Chen","suffix":""},{"id":565352664,"identity":"803d7670-e414-4421-9eef-821f07d49717","order_by":3,"name":"Shanglu Lin","email":"","orcid":"","institution":"Second Affiliated Hospital of Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Shanglu","middleName":"","lastName":"Lin","suffix":""},{"id":565352666,"identity":"e5fbdcfa-9eb5-41bb-9afe-e331d6eb3b24","order_by":4,"name":"Gang Wang","email":"","orcid":"","institution":"The Second Hospital \u0026 Clinical, Medical School, Lanzhou University","correspondingAuthor":false,"prefix":"","firstName":"Gang","middleName":"","lastName":"Wang","suffix":""},{"id":565352667,"identity":"d312e3d7-e3de-4e0e-840c-ea9642feaca8","order_by":5,"name":"Jinzhong Huang","email":"","orcid":"","institution":"Second Affiliated Hospital of Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Jinzhong","middleName":"","lastName":"Huang","suffix":""},{"id":565352669,"identity":"2883f2fa-fe37-4e3f-8f5c-78599798974a","order_by":6,"name":"Weizhi Qiu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAw0lEQVRIiWNgGAWjYFACHgaGhAoJHn5m5sMPiNfy4YyFjGQ7W5oB0VoYZ7ZU2Bic51GQIEqDwfmzB2/zNkjwGB/mYTBgqLGJJqzlwLlka94dEjxmh3kPPGA4lpbbQEiL2cEeM2neMyAtfAkGjA2HidBymAeopQ3osGYeAwnitBzjMZOcCdRiwEysFvszPMYWH4AOkzgMDOQEYvwi2X/G8EZCRZ09f//hww8+1NgQ1gICiOhIIEY5qpZRMApGwSgYBdgAADUnOlxuXKigAAAAAElFTkSuQmCC","orcid":"","institution":"Second Affiliated Hospital of Fujian Medical University","correspondingAuthor":true,"prefix":"","firstName":"Weizhi","middleName":"","lastName":"Qiu","suffix":""}],"badges":[],"createdAt":"2025-11-22 00:23:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8177019/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8177019/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":99217496,"identity":"5b781fb2-9a9c-43f2-9dfe-137d176edc0f","added_by":"auto","created_at":"2025-12-30 09:10:41","extension":"tif","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2506356,"visible":true,"origin":"","legend":"","description":"","filename":"figure1.tif","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/cf5d5b0795d82e623c1741ba.tif"},{"id":99217479,"identity":"e19c0304-6629-4d58-9767-ed6f2bf08b41","added_by":"auto","created_at":"2025-12-30 09:10:40","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1104618,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/d99462be3f2223c13bbf3574.docx"},{"id":99217497,"identity":"1503f178-6c39-45f8-91f2-23a6c89057bc","added_by":"auto","created_at":"2025-12-30 09:10:41","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":17538,"visible":true,"origin":"","legend":"","description":"","filename":"table1.docx","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/c16a2c9f1777657cb6796f0c.docx"},{"id":99318054,"identity":"5a6bcce1-6739-4502-ab73-de36dd26de33","added_by":"auto","created_at":"2025-12-31 16:31:17","extension":"tif","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6618844,"visible":true,"origin":"","legend":"","description":"","filename":"figure2.tif","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/10995eaed4f500216a7ad9dd.tif"},{"id":99319758,"identity":"6c9e1d20-c7a5-4e84-af33-b0e9641f0262","added_by":"auto","created_at":"2025-12-31 16:37:48","extension":"tif","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4485540,"visible":true,"origin":"","legend":"","description":"","filename":"figure3.tif","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/8f39a8f94d83f1a7e24075e7.tif"},{"id":99217509,"identity":"d9c2f956-6a05-4922-85cd-ea5d349b4952","added_by":"auto","created_at":"2025-12-30 09:10:42","extension":"tif","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3607776,"visible":true,"origin":"","legend":"","description":"","filename":"figure4.tif","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/5dcfbe2cc8612a666fbf2a75.tif"},{"id":99217480,"identity":"222cf521-29f2-4d51-9e41-afaa91d2093f","added_by":"auto","created_at":"2025-12-30 09:10:40","extension":"tif","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6692456,"visible":true,"origin":"","legend":"","description":"","filename":"figure5.tif","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/4168dfba8ce22ee39643bbb5.tif"},{"id":99217504,"identity":"47172174-8aea-4bee-bad5-e5cad6769f79","added_by":"auto","created_at":"2025-12-30 09:10:41","extension":"tif","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2054260,"visible":true,"origin":"","legend":"","description":"","filename":"figure6.tif","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/2eb83053ad101705943a323b.tif"},{"id":99317227,"identity":"0040bcc9-4af5-48b6-b2bb-51fd7583c8b8","added_by":"auto","created_at":"2025-12-31 16:29:48","extension":"json","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9551,"visible":true,"origin":"","legend":"","description":"","filename":"1ca88c7715474ad59969f3a7003ecaba.json","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/5109c3ae502ad061ec688fb8.json"},{"id":99217493,"identity":"5599f0b6-b1f0-42b1-9a2a-46533bf783ab","added_by":"auto","created_at":"2025-12-30 09:10:41","extension":"docx","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":19448,"visible":true,"origin":"","legend":"","description":"","filename":"suppplementtable1.docx","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/1aad16486303954972ded215.docx"},{"id":99217521,"identity":"14e4b6ca-86b3-4bc4-9f88-ea66ea1c3d3e","added_by":"auto","created_at":"2025-12-30 09:10:42","extension":"xml","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":153437,"visible":true,"origin":"","legend":"","description":"","filename":"1ca88c7715474ad59969f3a7003ecaba1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/c9db28107076f1cee3fa0d42.xml"},{"id":99318400,"identity":"dab53f69-edcb-4273-84d5-19de5dfe8ee1","added_by":"auto","created_at":"2025-12-31 16:33:04","extension":"tif","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2506356,"visible":true,"origin":"","legend":"","description":"","filename":"figure1.tif","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/04fa91db321ef4499667149c.tif"},{"id":99318380,"identity":"abc3e829-3f3f-4e64-beb9-aac0cdbfb7ab","added_by":"auto","created_at":"2025-12-31 16:32:57","extension":"tif","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6618844,"visible":true,"origin":"","legend":"","description":"","filename":"figure2.tif","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/112402f0104a77f151a10a5a.tif"},{"id":99217500,"identity":"9f2951c6-b86a-4701-ab56-75b46298fce5","added_by":"auto","created_at":"2025-12-30 09:10:41","extension":"tif","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4485540,"visible":true,"origin":"","legend":"","description":"","filename":"figure3.tif","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/ddd4977de98a31d5d9a68950.tif"},{"id":99217486,"identity":"d6a3f6ac-8172-4e79-8b60-7759b3ab0a9e","added_by":"auto","created_at":"2025-12-30 09:10:41","extension":"tif","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3607776,"visible":true,"origin":"","legend":"","description":"","filename":"figure4.tif","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/2b463bbc1508dfd4220d3a49.tif"},{"id":99217520,"identity":"eaa48190-5708-4f8f-8d0d-e65c355bf781","added_by":"auto","created_at":"2025-12-30 09:10:42","extension":"tif","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6692456,"visible":true,"origin":"","legend":"","description":"","filename":"figure5.tif","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/53553a04839307aaec71d592.tif"},{"id":99217503,"identity":"ceb47759-3f8e-4ef9-ab5b-8ffb2cb1ee88","added_by":"auto","created_at":"2025-12-30 09:10:41","extension":"tif","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2054260,"visible":true,"origin":"","legend":"","description":"","filename":"figure6.tif","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/8e3abcb68e17df2ad1c47b8b.tif"},{"id":99217514,"identity":"857e1d2b-7694-4e48-983c-22a61d389314","added_by":"auto","created_at":"2025-12-30 09:10:42","extension":"jpeg","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":208670,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/f330b2af90bfcbe24ebce87f.jpeg"},{"id":99217506,"identity":"894c5f7d-43ba-4bd6-bf84-72b8ef12987a","added_by":"auto","created_at":"2025-12-30 09:10:41","extension":"jpeg","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":261618,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/496ae7f549b2070f4cc0c4b1.jpeg"},{"id":99317065,"identity":"d1392c56-8d1a-461b-acb6-7da94915ef16","added_by":"auto","created_at":"2025-12-31 16:29:37","extension":"jpeg","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":443388,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/ad16f0b887ba1b2e557fadb7.jpeg"},{"id":99217518,"identity":"bdcd7863-ee17-46f6-a01b-5d629f1125b6","added_by":"auto","created_at":"2025-12-30 09:10:42","extension":"jpeg","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":201892,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/f0570d74962fd817c5e26ac3.jpeg"},{"id":99318180,"identity":"f0e6acc9-b8cc-4be6-a2c7-72d76b9d7cba","added_by":"auto","created_at":"2025-12-31 16:31:49","extension":"jpeg","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":189548,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/a99960fa920b4125139572ef.jpeg"},{"id":99318113,"identity":"263c8e40-650f-485a-90db-f7c94f0b4bef","added_by":"auto","created_at":"2025-12-31 16:31:32","extension":"jpeg","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":128388,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/e1cfc5aa187ef2b32c5d6432.jpeg"},{"id":99217515,"identity":"3e050e6e-7c80-488d-bc7a-c064737dd251","added_by":"auto","created_at":"2025-12-30 09:10:42","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":615639,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefigure1.png","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/177794da78a32b494c42113d.png"},{"id":99317443,"identity":"82f3e475-e514-4e75-b83a-c67d7402b3eb","added_by":"auto","created_at":"2025-12-31 16:30:14","extension":"png","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2137324,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefigure2.png","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/6c825f82a6baf0ecd9a31848.png"},{"id":99319733,"identity":"a5f8249e-bfdc-44a3-bb7b-6f14e97c91bc","added_by":"auto","created_at":"2025-12-31 16:37:44","extension":"png","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1455466,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefigure3.png","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/44bfb9cf9313ab6fc7928d90.png"},{"id":99317380,"identity":"694ea901-3dcc-43bf-a419-dcd6283c05a4","added_by":"auto","created_at":"2025-12-31 16:30:05","extension":"png","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1211081,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefigure4.png","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/77e9159f6035d6d2b4ef65c5.png"},{"id":99217482,"identity":"dba905a3-952b-49a8-a030-3fd9cdd6e645","added_by":"auto","created_at":"2025-12-30 09:10:40","extension":"png","order_by":27,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2711417,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefigure5.png","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/1e2f677804a9b2c468e042b3.png"},{"id":99318123,"identity":"bf5dfa01-c936-4280-b972-827dbcb87da3","added_by":"auto","created_at":"2025-12-31 16:31:35","extension":"png","order_by":28,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":604104,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefigure6.png","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/d4733134e185b89754ea79ee.png"},{"id":99217495,"identity":"807d8c86-020e-4bb2-8d49-43058e4c799d","added_by":"auto","created_at":"2025-12-30 09:10:41","extension":"png","order_by":29,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":49148,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/d075eddd046dfe2c5ea472c9.png"},{"id":99317042,"identity":"33c54b9a-c4f3-4703-921e-59cdf2dc7cb0","added_by":"auto","created_at":"2025-12-31 16:29:37","extension":"png","order_by":30,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":39853,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/b2ff9da5f2dac2b8c8f1e76a.png"},{"id":99317912,"identity":"3c803ea5-6929-4cc4-bbd4-fd00d2939a88","added_by":"auto","created_at":"2025-12-31 16:30:56","extension":"png","order_by":31,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":55473,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/d33de752a00347dd1660eb17.png"},{"id":99317691,"identity":"65714197-08b5-4624-9664-e7f7cd4d89f5","added_by":"auto","created_at":"2025-12-31 16:30:36","extension":"png","order_by":32,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":26816,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/02d89292148019fe207bd786.png"},{"id":99217524,"identity":"f470b02e-801d-4277-a00b-5bc86ab41926","added_by":"auto","created_at":"2025-12-30 09:10:42","extension":"png","order_by":33,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":31033,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/29ee862ec4c1f65b20e88901.png"},{"id":99217523,"identity":"e35fe4bd-c4c2-4d0d-bbee-3d0b3436a92c","added_by":"auto","created_at":"2025-12-30 09:10:42","extension":"png","order_by":34,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":18818,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/551c8753d2c8006101735fd4.png"},{"id":99217517,"identity":"0ba8d2d3-f1f6-4959-9b17-5ff6023929fc","added_by":"auto","created_at":"2025-12-30 09:10:42","extension":"xml","order_by":35,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":154524,"visible":true,"origin":"","legend":"","description":"","filename":"1ca88c7715474ad59969f3a7003ecaba1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/d9cd18b4cdfa9ada7162e7e7.xml"},{"id":99317035,"identity":"08eabedf-c5ad-4ad1-94ce-79e713f75b64","added_by":"auto","created_at":"2025-12-31 16:29:37","extension":"html","order_by":36,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":170609,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/559df4f8ae8becafcdd0d807.html"},{"id":99217476,"identity":"55d384d1-103b-4d01-9d43-da075d36156e","added_by":"auto","created_at":"2025-12-30 09:10:40","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":2214746,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlowchart of Patient Selection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 1. Flowchart of patient enrollment and exclusion\u003c/p\u003e\n\u003cp\u003eA total of 1,394 patients diagnosed with intracerebral hemorrhage (ICH) across two tertiary medical centers were initially screened. Among them, 735 individuals younger than 65 years of age were excluded, resulting in 659 older adults eligible for further review. Additional exclusions included: patients who were not admitted for the first time (N=120); those with concurrent malignant tumors (N=35); those with non-initial hospitalizations (N=58); and those who underwent tracheotomy more than 7 days after admission (N=11). Following these criteria, 435 elderly patients with ICH were ultimately included in the final analysis cohort.\u003c/p\u003e","description":"","filename":"figure1.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/77ed885b694c0027500ccf21.jpg"},{"id":99217489,"identity":"3c83c83f-07a4-4d19-acdb-82da9409d570","added_by":"auto","created_at":"2025-12-30 09:10:41","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":3597924,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eLASSO-Based Feature Selection and Variable Contribution Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 2 Feature selection and variable contribution using the LASSO regression model\u003c/p\u003e\n\u003cp\u003e(A) Association between the regularization parameter (alpha) and the number of non-zero coefficients. As alpha increases, coefficients shrink toward zero, and fewer features remain in the model; the optimal alpha (0.0118) is indicated with a vertical dashed red line.\u003c/p\u003e\n\u003cp\u003e(B) LASSO coefficient path plot. Each line represents a feature coefficient trajectory as alpha varies on a logarithmic scale. The vertical dashed red line marks the optimal alpha chosen through ten-fold cross-validation, corresponding to maximal model stability with minimal mean squared error.\u003c/p\u003e\n\u003cp\u003e(C) Selected feature importance based on LASSO coefficients. Five predictors retained after LASSO regularization are shown ranked by absolute coefficient magnitude: GCS, serum albumin, midline shift, IVH, and CT biphasic sign.\u003c/p\u003e\n\u003cp\u003e(D) Correlation heatmap of selected features. All pairwise Pearson correlation coefficients are below 0.6, indicating absence of significant multicollinearity among the chosen predictors.\u003c/p\u003e\n\u003cp\u003e(E) Residual distribution plot. Residuals from the LASSO model follow an approximately normal distribution, suggesting good model fit and statistical stability.\u003c/p\u003e\n\u003cp\u003e(F) Standardized density distributions of selected features. The five core predictors show relatively comparable distribution patterns, supporting their suitability for downstream machine learning modeling.\u003c/p\u003e","description":"","filename":"figure2.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/b997c96389e01590115688e2.jpg"},{"id":99217507,"identity":"45387bff-9124-44d1-aaba-ef5424cb3ad0","added_by":"auto","created_at":"2025-12-30 09:10:41","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":2398239,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePerformance Comparison of Six ML Models\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 3 Comparative performance of six machine learning models in predicting early tracheotomy in elderly patients with ICH.\u003c/p\u003e\n\u003cp\u003e(A) Composite performance scores of all six algorithms—Random Forest (RF), XGBoost, Extra Trees, Support Vector Machine (SVM), Gradient Boosting, and Logistic Regression—demonstrating the overall superiority of the RF model.\u003c/p\u003e\n\u003cp\u003e(B) Radar chart comparing accuracy, precision, recall, F1 score, and AUC across models, with RF showing the most balanced and robust performance profile.\u003c/p\u003e\n\u003cp\u003e(C) ROC curves displaying discrimination ability for each model; RF achieves the highest AUC (0.687) among all methods.\u003c/p\u003e\n\u003cp\u003e(D) Precision–recall (PR) curves illustrating performance in the minority (tracheotomy) class; the RF model provides the best trade-off between precision and recall.\u003c/p\u003e\n\u003cp\u003e(E) Bar chart showing the numerical values of core evaluation metrics across all models, again highlighting the consistently stronger performance of Random Forest.\u003c/p\u003e","description":"","filename":"figure3.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/17cf36ba13359ea9ff2f5bd1.jpg"},{"id":99217501,"identity":"b6b929fb-aee8-4015-99b0-2f93c728b2ce","added_by":"auto","created_at":"2025-12-30 09:10:41","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1941431,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePerformance of the Random Forest Model After Hyperparameter Optimization\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 4 Evaluation of the optimized Random Forest (RF) model following Bayesian hyperparameter tuning.\u003c/p\u003e\n\u003cp\u003e(A) ROC curve demonstrating improved discrimination after tuning, with an AUC of 0.723 in the test set.\u003c/p\u003e\n\u003cp\u003e(B) Precision–recall (PR) curve showing a PR-AUC of 0.447, with preserved precision at increasing recall levels—important for minority-class detection.\u003c/p\u003e\n\u003cp\u003e(C) Calibration curve comparing predicted versus observed probabilities; the curve closely follows the diagonal, indicating good probability calibration.\u003c/p\u003e\n\u003cp\u003e(E) Radar chart summarizing overall model performance across accuracy, precision, recall, F1 score, and AUC, illustrating the balanced nature of the optimized RF model.\u003c/p\u003e\n\u003cp\u003e(F) Heatmap of detailed performance metrics, showing accuracy (0.793), precision (0.784), recall (0.793), F1 score (0.788), and AUC (0.723), confirming the model's stability and robustness.\u003c/p\u003e","description":"","filename":"figure4.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/5d3670836e85d85c9a9e0de4.jpg"},{"id":99217510,"identity":"6ce4bbb8-34d1-4eb8-a180-b5d1df070b93","added_by":"auto","created_at":"2025-12-30 09:10:42","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":3216866,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP-Based Interpretability Analysis of the Final Model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 5 Model interpretability analysis using SHapley Additive exPlanations (SHAP).\u003c/p\u003e\n\u003cp\u003e(A) Global SHAP beeswarm plot showing the distribution of SHAP values for each feature; lower GCS score, IVH, midline shift, and lower albumin strongly push predictions toward higher tracheotomy risk.\u003c/p\u003e\n\u003cp\u003e(B) SHAP feature importance ranking based on mean absolute SHAP values, identifying GCS as the dominant predictor, followed by IVH, midline shift, albumin, and CT biphasic sign.\u003c/p\u003e\n\u003cp\u003e(C) SHAP decision plot visualizing how features sequentially steer individual predictions toward high- or low-risk classifications.\u003c/p\u003e\n\u003cp\u003e(D) SHAP waterfall plot for a tracheotomy (TT) patient, highlighting risk-increasing features (red) and protective features (blue).\u003c/p\u003e\n\u003cp\u003e(E) SHAP waterfall plot for a non-tracheotomy (Non-TT) patient, illustrating how the balance of feature contributions shifts the prediction toward non-intervention. In both plots, f(x) represents the model output, and E[f(x)] is the baseline expected value.\u003c/p\u003e","description":"","filename":"figure5.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/774d25370a377d3015c7c91f.jpg"},{"id":99217499,"identity":"d0619afe-ac56-45aa-94d1-e3838c8e6a79","added_by":"auto","created_at":"2025-12-30 09:10:41","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":1135783,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eWeb-Based Tracheotomy Prediction Platform\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 6 Online prediction platform for estimating tracheotomy risk using the optimized Random Forest model.\u003c/p\u003e\n\u003cp\u003eShown is the interface of the deployed web-based tool (http://47.109.146.0:8000/), where clinicians input five key clinical variables to obtain a real-time prediction of tracheotomy probability.\u003c/p\u003e\n\u003cp\u003eUsing a representative patient with GCS of 15, albumin 34.3 g/L, midline shift of 4 mm, CT blend sign, and intraventricular hemorrhage, the system outputs a predicted tracheotomy probability of 76.64%, with a 95% confidence interval of [68.62%, 84.65%], displayed instantly on the interface.\u003c/p\u003e","description":"","filename":"figure6.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/5a064b8b57a4938709b272cc.jpg"},{"id":108181598,"identity":"9a406dc3-5d90-40da-b150-233d051add5c","added_by":"auto","created_at":"2026-04-30 08:58:46","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":14900432,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/f07d89ec-8d32-414c-af0a-ecb09fd1fe14.pdf"},{"id":99217477,"identity":"b6b1d674-4069-4dba-9184-08d99112884e","added_by":"auto","created_at":"2025-12-30 09:10:40","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":19448,"visible":true,"origin":"","legend":"","description":"","filename":"suppplementtable1.docx","url":"https://assets-eu.researchsquare.com/files/rs-8177019/v1/5192fd0f052ae0596fe06c83.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"\u003cp\u003eMachine learning–based risk stratification for early tracheotomy in geriatric intracerebral hemorrhage: model development and web deployment\u003c/p\u003e","fulltext":[{"header":"Introduction","content":"\u003cp\u003eIntracerebral hemorrhage (ICH) is one of the most severe causes of stroke and is associated with the highest mortality and disability rates among the elderly. In older patients, less physiological reserve, decreased compensation ability, and multiple comorbidities may result in worse neurological outcomes than in younger patients[1, 2]. Aspiration pneumonia, airway obstruction, and hypoventilation, respiratory complications occur more than 60% of ICH patients and are an important cause of secondary hypoxemia[3]. Hypoxemia can aggravate brain tissue hypoxia, exacerbate perihematomal edema, and lead to adverse changes in intracranial pressure (ICP), ultimately resulting in secondary brain injury[4]. The above complications are more common in the elderly. Weakened cough reflex, airway protection ability, and lung compliance, leading to a high incidence of respiratory failure[5]. Timely and appropriate management of airway protection is critical to prevent secondary brain injury and improve outcomes of elderly ICH patients.\u003c/p\u003e \u003cp\u003eTracheotomy (TT) is a common surgical procedure to provide airway protection, improve ventilation, prevent aspiration, and reduce hypoxemia in patients with severe neurological deficits[6]. TT can shorten sedation time, facilitate pulmonary toileting, and reduce the incidence of ventilator-associated pneumonia and other complications when performed early[7]. However, the optimal timing for tracheotomy has not been well established and may vary depending on the attending physician\u0026rsquo;s experience and clinical judgment. There is no widely accepted consensus or standardized timing window for TT, leading to delayed treatment, respiratory deterioration, and secondary neurological injury[8, 9]. Although several clinical predictors, such as Glasgow Coma Scale (GCS) score, hematoma volume, and brainstem compression, have been proposed to guide clinical decision-making, there is no validated tool to help clinicians identify ICH patients who would benefit from TT the most. Studies specifically evaluating these predictors in the elderly ICH population are even more serious.\u003c/p\u003e \u003cp\u003eAlthough several previous studies have evaluated the risk factors of TT and tried to determine the optimal timing[10\u0026ndash;12], there were some important limitations that should be taken into consideration. The majority of studies were from single centers and had relatively small samples without external validation or more diverse patient cohorts[11]. Some studies used traditional linear or logistic regression models, which were suboptimal for capturing complex nonlinear relationships between features and the target. More importantly, most of the studies did not focus on older ICH patients, who are more vulnerable to airway complications but also have different clinical characteristics. To overcome these limitations, we performed a multicenter analysis of elderly ICH patients and utilized several machine learning (ML) methods including LASSO feature selection, Random Forest (RF) classification, Bayesian optimization, and SHAP interpretability. Compared with previous studies, this design can lead to more accurate prediction, better robustness, and improved interpretability for clinical use. We also developed and implemented a web-based tool for clinical use based on the prediction model, which could serve as an evidence-based clinical decision support system to help clinicians identify ICH patients with high risk for tracheotomy, particularly for the elderly.\u003c/p\u003e \u003cp\u003e This study was a multicenter, retrospective cross-sectional investigation reported in accordance with the STROBE guidelines. It aimed to explore the association between clinical characteristics, relevant indicators, and outcomes\u0026mdash;specifically whether elderly patients with ICH underwent tracheotomy.\u003c/p\u003e"},{"header":"Methods and Materials","content":"\u003cp\u003e\u003cstrong\u003eStudy Population\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study population was drawn from two tertiary teaching hospitals in China. Specifically, it included patients admitted to the Second Hospital of Lanzhou University between March 2022 and December 2023, and patients admitted to the Second Affiliated Hospital of Fujian Medical University between January 2015 and April 2022 who were diagnosed with ICH. A total of 1,394 patients were initially screened, among whom 435 elderly patients aged\u0026nbsp;\u0026ge;65 years met the eligibility criteria and were included as the final analytical cohort. All enrolled patients were diagnosed with spontaneous ICH based on the International Classification of Diseases, Ninth Revision (ICD-9) or Tenth Revision (ICD-10), according to standard diagnostic criteria.\u003c/p\u003e\n\u003cp\u003eThe exclusion criteria were as follows: 1) Age \u0026lt;65 years; 2) Secondary ICH due to trauma, brain tumors, aneurysms, or arteriovenous malformations; 3) Patients with multiple hospitalizations for ICH, for whom only the first admission was considered; 4) Severe hepatic or renal dysfunction, leukemia, lymphoma, other hematologic disorders, or malignancies; and 5) Incomplete laboratory data. Ultimately, 435 elderly patients who met all inclusion criteria were enrolled in the study (Figure 1).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eEthical approval was obtained from the Ethics Committee of\u0026nbsp;the Second Hospital \u0026amp; Clinical, Medical School, Lanzhou University (Approval No. 2025A-1073) and the Ethics Committee of the Second Affiliated Hospital of Fujian Medical University (Approval No. 2024-016). The study strictly adhered to the ethical principles of the Declaration of Helsinki. All research data were anonymized and de-identified before use, and the requirement for written informed consent was waived by the ethics committees.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eVariable Collection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA comprehensive set of predictor variables associated with the outcome of TT was collected based on prior literature, including demographic characteristics, lifestyle factors, comorbidities, clinical symptoms, neuroimaging features, and laboratory indicators. Demographic variables consisted of age and sex; lifestyle factors included current smoking and alcohol consumption. Comorbidities relevant to intracerebral hemorrhage, such as hypertension and diabetes mellitus, were also recorded. Clinical status variables included GCS score, body temperature, respiratory rate, heart rate, systolic and diastolic blood pressure. Neuroimaging variables included the location of hemorrhage (lobar, basal ganglia, thalamus, cerebellum, or brainstem), presence of intraventricular hemorrhage (IVH), CT hematoma signs (none, island sign, blend sign, fluid level sign, or black hole sign), laterality of the hematoma, midline shift, and initial hematoma volume. Hematoma volume was calculated by radiologists using manual segmentation in 3D Slicer software. Laboratory data were obtained from the patients\u0026rsquo; initial admission tests. These included glucose, uric acid, electrolytes (potassium, sodium, calcium, phosphorus, magnesium), total protein, albumin, complete blood count parameters (white blood cell count, neutrophils, lymphocytes, monocytes, hemoglobin, platelets), and coagulation indices (prothrombin time [PT], international normalized ratio [INR], activated partial thromboplastin time [APTT], fibrinogen, thrombin time [TT], and D-dimer).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDefinition of Outcome\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe primary outcome of this study was early TT, defined as the performance of a TT within 7 days after hospital admission[13, 14]. Patients who did not undergo TT or received the procedure after day 7 were classified into the non-TT group.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Processing and ML Modeling\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll patient samples were randomly divided into a training set and a testing set in a 7:3 ratio. Six supervised machine learning algorithms were employed for model development, including logistic regression (LR), RF, extreme gradient boosting (XGBoost), support vector machine (SVM), K-nearest neighbors (KNN), and an ensemble model. These algorithms were selected to balance linear and nonlinear modeling capabilities and to compare performance in terms of structural complexity, interpretability, and computational efficiency. During preprocessing, missing values of continuous variables were imputed using the median, while categorical variables were transformed into numerical form through one-hot encoding to ensure compatibility with all models. All variables were subsequently standardized to minimize potential bias due to differences in measurement scales. Preliminary exploratory analysis detected no abnormal outliers, and the distribution and scale of features were generally consistent, supporting their suitability for model training.\u003c/p\u003e\n\u003cp\u003eFeature selection was performed using the least absolute shrinkage and selection operator (LASSO) regression, with ten-fold cross-validation used to determine the optimal regularization parameter \u0026lambda;. A coefficient trace plot was generated to visually demonstrate the feature shrinkage process. Considering the imbalance in the original dataset (with relatively few patients receiving tracheotomy), we applied the Synthetic Minority Oversampling Technique (SMOTE) together with class weight adjustments during model training to mitigate bias toward the majority class.\u003c/p\u003e\n\u003cp\u003eHyperparameters for each algorithm were optimized using Bayesian optimization embedded within five-fold cross-validation, which enhances model stability and generalization while reducing redundant parameter search and computational cost. The final optimal hyperparameters were: max_depth = 17, max_features = \u0026lsquo;log2\u0026rsquo;, min_samples_leaf = 1, min_samples_split = 2, and n_estimators = 107. Comprehensive model performance was evaluated using receiver operating characteristic (ROC) curves and area under the curve (AUC), precision\u0026ndash;recall (PR) curves, calibration curves, as well as accuracy, precision, recall, and F1 score.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel Interpretation and Web Deployment\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo address the \u0026ldquo;black-box\u0026rdquo; limitation commonly associated with traditional machine learning models, we incorporated Shapley Additive Explanations (SHAP) to quantify the contribution of each feature to the prediction and to enhance interpretability. SHAP provides both global and individual-level explanations, allowing us not only to understand overall feature importance trends but also to interpret predictions for specific patients in a personalized manner.\u003c/p\u003e\n\u003cp\u003eTo further improve clinical usability and operational convenience, we deployed the optimized predictive model as an online application using the Flask framework. Clinicians can obtain a real-time prediction of a patient\u0026apos;s tracheotomy risk by simply entering the required key feature values. Importantly, the prediction platform has been made freely accessible to both healthcare professionals and patients, thereby enhancing the translational potential and real-world applicability of the model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistical Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCategorical variables were compared between groups using the chi-square test or Fisher\u0026rsquo;s exact test, as appropriate. Continuous variables were analyzed using the independent samples t-test or the Mann\u0026ndash;Whitney U test based on the results of normality testing. Model performance was assessed using multiple evaluation metrics, including the AUC, accuracy, precision, recall, and F1 score. The AUC values of the training and testing sets were used as the primary indicators of predictive performance. A two-sided p-value \u0026le; 0.05 was considered statistically significant. All data processing and statistical analyses were conducted within a conda virtual environment in PyCharm.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eBaseline Characteristics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA total of 435 elderly patients with ICH were included in this study, of whom 83 underwent TT (TT group) and 352 did not (Non-TT group), as shown in Table 1. Baseline comparisons revealed significant differences between the two groups across multiple variables, which may influence or reflect the risk of requiring TT.\u003c/p\u003e\n\u003cp\u003ePatients in the TT group exhibited more severe neurological impairment, including lower GCS scores, larger initial hematoma volumes, and more pronounced midline shift. In addition, inflammatory markers were elevated in the TT group, with higher white blood cell and neutrophil counts, whereas lymphocyte counts were lower compared with the Non-TT group. Blood glucose levels were also significantly higher in the TT group. Regarding underlying comorbidities, hypertension was more prevalent in the TT group. The biphasic CT sign appeared more frequently as well. Moreover, basal ganglia hemorrhage and IVH were markedly more common in the TT group, indicating that these patients tended to have more complex and severe disease, which may predispose them to requiring TT.\u003c/p\u003e\n\u003cp\u003eThe dataset was divided into training and testing sets at a 7/3 ratio, and comparison showed good balance between the two subsets (Supplementary Table 1).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFeature Selection and Variable Contribution\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo initiate the modeling process, feature selection was performed using all patient data within the training set. Key predictors were identified through the least absolute shrinkage and selection operator (LASSO) regression, with ten-fold cross-validation used to determine the optimal regularization parameter \u0026alpha;. The LASSO coefficient path plot (Figure 2A) illustrates the dynamic shrinkage of feature coefficients as \u0026alpha; changes. The optimal \u0026alpha; value of 0.0118 (Figure 2B) was selected based on the minimum mean squared error (MSE), yielding a stable and high-performing model while effectively reducing multicollinearity. A total of five significant predictors were retained after LASSO screening: GCS score, IVH, midline shift, serum albumin level, and the CT blend sign (Figure 2C). These features were subsequently used for model development. Correlation analysis (Figure 2D) showed that all pairwise correlation coefficients were below 0.6, suggesting no considerable multicollinearity among the selected variables. Residual distribution analysis (Figure 2E) demonstrated an approximately normal distribution, supporting the good fit and statistical stability of the LASSO model. Furthermore, density distribution plots indicated that the five key features showed consistent patterns in value distribution and central tendency, providing a solid foundation for exploring their relationships with disease phenotypes and clinical outcomes.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eML Model Testing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo predict the risk of tracheotomy among elderly patients with ICH, we evaluated six supervised machine learning algorithms: RF, XGBoost, Extra Trees, SVM, Gradient Boosting, and Logistic Regression. The performance of each model was assessed across multiple metrics, including AUC, accuracy, precision, recall, and F1 score. Visualization was conducted using comprehensive scoring plots, radar charts, ROC curves, PR curves, and bar charts. Among all models, the RF algorithm demonstrated the best overall performance. In terms of composite scores, the RF model achieved the highest value (0.831), significantly outperforming the other algorithms (Figure 3A). The ROC curve showed that Random Forest had the largest AUC (0.687), surpassing XGBoost (0.647) and SVM (0.658) (Figure 3C). The PR curve further revealed that Random Forest maintained a favorable balance between precision and recall compared with other models (Figure 3D). Radar charts (Figure 3B) and bar plots (Figure 3E) consistently confirmed that the Random Forest model ranked highest in accuracy, recall, and F1 score.\u003c/p\u003e\n\u003cp\u003eIn summary, these findings indicate that RF provided the most robust and reliable predictive performance among the six tested algorithms and was therefore selected as the core model for subsequent optimization and deployment.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel Optimization and Evaluation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo further enhance predictive performance and generalizability, Bayesian optimization was employed to automatically tune the key hyperparameters of the model. Using ten-fold cross-validation with the average ROC\u0026ndash;AUC as the objective function, the algorithm efficiently identified the optimal parameter combination: max_depth = 17, max_features = \u0026ldquo;log2\u0026rdquo;, min_samples_leaf = 1, min_samples_split = 2, and n_estimators = 107.\u003c/p\u003e\n\u003cp\u003eAfter determining the optimal parameters, the final RF model was retrained on the full training set and subsequently evaluated on the independent testing set. The ROC curve demonstrated improved predictive accuracy after tuning, with an AUC of 0.723 (Figure 4A). The precision\u0026ndash;recall curve showed a PR-AUC of 0.447, and precision gradually decreased as recall increased, indicating that the model preserved good balance in minority-class prediction, which is crucial for identifying high-risk individuals (Figure 4B). The calibration curve revealed strong agreement between predicted probabilities and observed outcomes, closely approximating the ideal diagonal line and suggesting reliable probability estimation (Figure 4C).\u003c/p\u003e\n\u003cp\u003eAdditional performance metrics showed that the optimized Random Forest model achieved balanced results across all indices (Figure 4D). The heatmap (Figure 4E) summarized the key performance metrics: accuracy (0.793), precision (0.784), recall (0.793), F1 score (0.788), and AUC (0.723), collectively demonstrating the model\u0026rsquo;s stability, robustness, and potential for clinical application in predicting tracheotomy risk.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSHAP-Based Model Interpretability Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo enhance the clinical interpretability of the predictive model, we applied SHapley Additive exPlanations (SHAP) to systematically analyze the contribution of each key feature to tracheotomy risk. The global SHAP beeswarm plot (Figure 5A) demonstrated that the GCS score was the most influential factor, with lower GCS values associated with markedly increased SHAP values, indicating a higher likelihood of tracheotomy. IVH and midline shift also contributed positively to risk prediction, reflecting the importance of structural brain injury. Lower serum albumin levels were associated with increased risk, whereas the CT blend sign had a relatively smaller but still notable effect. The SHAP feature importance ranking (Figure 5B) further confirmed the overall contributions of GCS, IVH, midline shift, albumin, and CT hematoma signs in descending order of importance. The SHAP decision plot (Figure 5C) illustrated how each feature sequentially influenced the predicted risk, consistent with the patterns observed in global explanations.\u003c/p\u003e\n\u003cp\u003eAdditionally, SHAP waterfall plots were generated for two representative cases: one patient who underwent tracheotomy (Figure 5D) and one who did not (Figure 5E). These plots clearly showed the positive (risk-increasing) and negative (risk-reducing) contributions of individual features to each patient\u0026rsquo;s prediction. In the figures, f(x) represents the model\u0026rsquo;s predicted output, and E[f(x)] represents the baseline expected value. Red bars denote risk-enhancing features, while blue bars represent protective factors, with bar length indicating the magnitude of influence.\u003c/p\u003e\n\u003cp\u003eOverall, the SHAP analysis substantially improved the transparency of the model\u0026rsquo;s decision-making process and supports personalized clinical risk assessment and decision-making.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eOnline Web Deployment\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 6 showed that to enhance the clinical applicability of the model, we developed and deployed an online tracheotomy risk prediction system based on the optimized RF algorithm using the Flask framework. The platform (http://47.109.146.0:8000/) allows users to obtain real-time predictions by simply entering five core clinical features. The system automatically generates the predicted probability of early TT for patients with ICH, thereby improving accessibility and convenience for clinical decision-making.\u003c/p\u003e\n\u003cp\u003eAs an example, for a representative patient with a GCS score of 15, serum albumin level of 34.3 g/L, midline shift of 4 mm, positive CT blend sign, and presence of intraventricular hemorrhage, the system estimated a tracheotomy probability of 76.64%, with a 95% confidence interval of [68.62%, 84.65%]. The predicted probability and confidence interval are displayed immediately on the right side of the interface, offering an intuitive and user-friendly tool for clinicians.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis multicenter cohort study among older adults with ICH developed and validated a ML\u0026ndash;based clinical prediction model for the support of early identification of ICH patients who would most likely benefit from early TT. Leveraging the integration of routinely available clinical, imaging, and laboratory variables, this model can offer more accurate and individualized risk stratification than experience-based clinical decision making. The practical clinical utility is that the web-based ML model can estimate the individualized TT risk at admission, and thus, if implemented, help clinicians in optimizing airway management in this patient population, avoiding secondary hypoxemia, and improving the outcomes among these patients.\u003c/p\u003e\n\u003cp\u003ePrevious models of predictors for TT or prolonged ventilation in neurocritical populations have several limitations. While earlier studies have established factors such as low GCS scores, intraventricular hemorrhage, and large hematoma burden as being more likely to undergo tracheotomy, these were mostly developed using cohorts with small sample sizes from single centers as these research by Shadi Yaghi et al. and Szeder et al.[10, 11]. Later models, including the commonly used SET score which was built by\u0026nbsp;Silvia et al.[15], attempted to incorporate neurological, radiological, and systemic variables and demonstrated moderate discrimination in stroke and ICH populations[12, 15]. However, external validations of these scores have consistently shown moderate discrimination at best, with variation between centers and patient subgroups. Older adults, a key focus of this study, are underrepresented in most previous prediction models of prolonged mechanical ventilation after ICH[16].\u003c/p\u003e\n\u003cp\u003eIn comparison with previous efforts, our study has several novel characteristics. First, we focused specifically on the ICH population in geriatric patients, a group which is at particularly high risk of developing respiratory complications after ICH yet to date lacks a dedicated prediction tool. Second, by using a ML approach rather than a predefined scoring system, we were able to capture nonlinear interactions between variables. Third, we provided feature explainability, and released the model as an online calculator to better support its clinical applicability in guiding more individualized airway-management decisions in older adults.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp; Predictors included in the RF model are also biologically plausible. Low GCS score suggests decreased level of consciousness and airway protective reflexes and can predispose to aspiration, hypoventilation, and hypoxemia[17]. IVH and midline shift indicate severe structural brain injury that may lead to a deeper sedation level and longer duration of mechanical ventilation[18, 19]. Hypoalbuminemia could suggest the presence of systemic inflammatory response and malnutrition or poor nutritional reserves in critically ill patients as well as a state of increased catabolism, which is a condition of high metabolic demand in the human body[20, 21]. Therefore, it can be a predisposing factor for poor weaning outcome. Biphasic CT sign was associated with hematoma heterogeneity and early hematoma rapid expansion, which could lead to indirect neurological compromise[22]. It is of note that most of these clinical and radiographic predictors of post-ICH respiratory deterioration were statistically significant and biologically plausible, as included in our model. ML has the potential to combine a number of multidimensional predictors of outcomes in an interpretable way that aligns with the actual thought process of clinical care in the real world more so than looking at individual parameters. In a geriatric medicine view, this model may help identify older adults at higher risk of post-ICH respiratory deterioration earlier than clinical judgment alone and thus will be able to determine whether the need for a TT is likely, which will hopefully allow earlier intervention and decreased secondary hypoxic injury and in-hospital death as well as higher functional outcome.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp; There are several clinically relevant takeaways from this work. First, we provide an evidence-based tool to inform clinicians about the risk of early TT for a given older patient. While physicians commonly make airway management decisions, those decisions and the timing of them can be highly subjective and variable between clinicians. A risk prediction model to aid in airway decision-making can help address such limitations and promote standardization, equity, and timeliness. This study hypothesize that knowing ahead of time the risk of early tracheotomy may also allow for earlier anticipation, family discussion, allocation of neurocritical care resources, and earlier respiratory care interventions that might limit secondary hypoxemia and preventable complications. Second, the risk prediction model produces individualized risk estimates, at the time of admission, that can aid in airway management decisions. These patient-level predictions can be instrumental in the growing field of precision geriatric medicine, in which intervention selection is individualized based on physiologic rather than chronological age. Third, this approach provides a user-friendly web-based application, which can be accessed at the bedside with no need for specialty software, that may promote real-world uptake and use in clinical care.\u003c/p\u003e\n\u003cp\u003eThis study has several limitations that need to be addressed. First, the data were collected from two large tertiary referral hospitals, and the retrospective design may limit the generalizability of the results and introduce selection bias. Clinical decision-making patterns may vary across institutions, and external validation in broader and more community-based geriatric populations will be necessary to ensure the robustness of the model. Second, while the model included important clinical, imaging, and laboratory variables, other potential predictors, such as frailty status, pre-morbid functional status, cognitive impairment, and detailed cardiopulmonary assessments, were not consistently available and could be further explored to enhance the predictive performance. Third, the ML methods still rely on the quality and completeness of the data, and their results may be affected by unmeasured confounders. Prospective studies with standardized data collection, predefined TT criteria, and a focus on assessing the clinical utility of the model are needed. Future research could also consider integrating longitudinal physiological data, continuous monitoring signals, and frailty-informed geriatric assessments. Finally, implementation studies are necessary to understand how the prediction tool can be integrated into clinical workflows, impact decision-making timing, and improve patient-centered outcomes.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn this multicenter cohort of older adults with intracerebral hemorrhage, the research developed and validated an interpretable ML model that predicted early TT using five easily obtained admission variables. The model had balanced performance and enhanced predictive accuracy compared with traditional clinical methods. By combining important neurological, imaging, and laboratory markers, and providing an easily accessible web-based calculator, this study offers a practical decision-support tool that may help clinicians identify high-risk patients earlier and refine airway management strategies. These results underscore the potential of data-driven approaches to facilitate personalized care and reduce secondary respiratory complications in geriatric neurocritical populations.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe thank the clinical teams at Lanzhou University Second Hospital and Fujian Medical University Second Affiliated Hospital for their support in data retrieval and verification. We also acknowledge the contributions of the patients and their families who made this research possible.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWJY (First Author), CLJ: (Co-first Author) Conceptualization, methodology, data curation, formal analysis, writing\u0026mdash;original draft. FZY (Co-first Author): Data collection, imaging review, statistical validation. LSL: Machine-learning modeling, software development, visualization. WG: Clinical interpretation, manuscript review, supervision. HJZ (Co-corresponding Author): Project administration, resource coordination, writing\u0026mdash;review \u0026amp; editing. QWZ (Corresponding Author): Study design, critical revision, final approval, and correspondence.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Ethics Committee of the Second Hospital \u0026amp; Clinical, Medical School, Lanzhou University (Approval No. 2025A-1073) and Second Affiliated Hospital of Fujian Medical University (Approval No. 2024-016). All methods adhered to the Declaration of Helsinki. Because this was a retrospective analysis of anonymized data, the requirement for written informed consent was waived.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHuman Ethics and Consent to Participate Declarations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical Trial Number\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eClinical trial number: not applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and analyzed during the current study are not publicly available due to institutional restrictions but are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eJoint funds for the innovation of science and technology, Fujian province(Grant number:2023Y9239）\u003c/p\u003e\n\u003cp\u003eGuiding Project of Quanzhou Science and Technology Bureau (Grant number: 2023N064S)\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eRibo M, Grotta JC. Latest advances in intracerebral hemorrhage. Curr Neurol Neurosci Rep. 2006;6:17\u0026ndash;22. https://doi.org/10.1007/s11910-996-0004-0.\u003c/li\u003e\n\u003cli\u003eGaist D, Wallander M-A, Gonz\u0026aacute;lez-P\u0026eacute;rez A, Garc\u0026iacute;a-Rodr\u0026iacute;guez LA. Incidence of hemorrhagic stroke in the general population: validation of data from the health improvement network. Pharmacoepidemiol Drug Saf. 2013;22:176\u0026ndash;82. https://doi.org/10.1002/pds.3391.\u003c/li\u003e\n\u003cli\u003eMaramattom BV, Weigand S, Reinalda M, Wijdicks EFM, Manno EM. Pulmonary complications after intracerebral hemorrhage. Neurocrit Care. 2006;5:115\u0026ndash;9. https://doi.org/10.1385/NCC:5:2:115.\u003c/li\u003e\n\u003cli\u003eYang Y, Rosenberg GA. Blood-brain barrier breakdown in acute and chronic cerebrovascular disease. Stroke. 2011;42:3323\u0026ndash;8. https://doi.org/10.1161/STROKEAHA.110.608257.\u003c/li\u003e\n\u003cli\u003eLu P, Cao Z, Gu H, Li Z, Wang Y, Cui L, et al. Association of sex and age with in-hospital mortality and complications of patients with intracerebral hemorrhage: a study from the chinese stroke center alliance. Brain Behav. 2023;13:e2846. https://doi.org/10.1002/brb3.2846.\u003c/li\u003e\n\u003cli\u003eB\u0026ouml;sel J, Niesen W-D, Salih F, Morris NA, Ragland JT, Gough B, et al. Effect of early vs standard approach to tracheostomy on functional outcome at 6 months among patients with severe stroke receiving mechanical ventilation: the SETPOINT2 randomized clinical trial. JAMA. 2022;327:1899\u0026ndash;909. https://doi.org/10.1001/jama.2022.4798.\u003c/li\u003e\n\u003cli\u003eDing W-L, Xiang Y-S, Liao J-C, Wang S-Y, Wang X-Y. Early tracheostomy is associated with better prognosis in patients with brainstem hemorrhage. J Integr Neurosci. 2020;19:437\u0026ndash;42. https://doi.org/10.31083/j.jin.2020.03.25.\u003c/li\u003e\n\u003cli\u003eKo S-B, Choi HA, Parikh G, Helbok R, Schmidt JM, Lee K, et al. Multimodality monitoring for cerebral perfusion pressure optimization in comatose patients with intracerebral hemorrhage. Stroke. 2011;42:3087\u0026ndash;92. https://doi.org/10.1161/STROKEAHA.111.623165.\u003c/li\u003e\n\u003cli\u003eImperiale C, Magni G, Favaro R, Rosa G. Intracranial pressure monitoring during percutaneous tracheostomy \u0026ldquo;percutwist\u0026rdquo; in critically ill neurosurgery patients. Anesth Analg. 2009;108:588\u0026ndash;92. https://doi.org/10.1213/ane.0b013e31818f601b.\u003c/li\u003e\n\u003cli\u003eSzeder V, Ortega-Gutierrez S, Ziai W, Torbey MT. The TRACH score: clinical and radiological predictors of tracheostomy in supratentorial spontaneous intracerebral hemorrhage. Neurocrit Care. 2010;13:40\u0026ndash;6. https://doi.org/10.1007/s12028-010-9346-1.\u003c/li\u003e\n\u003cli\u003eYaghi S, Moore P, Ray B, Keyrouz SG. Predictors of tracheostomy in patients with spontaneous intracerebral hemorrhage. Clin Neurol Neurosurg. 2013;115:695\u0026ndash;8. https://doi.org/10.1016/j.clineuro.2012.08.010.\u003c/li\u003e\n\u003cli\u003eAlsherbini K, Goyal N, Metter EJ, Pandhi A, Tsivgoulis G, Huffstatler T, et al. Predictors for tracheostomy with external validation of the stroke-related early tracheostomy score (SETscore). Neurocrit Care. 2019;30:185\u0026ndash;92. https://doi.org/10.1007/s12028-018-0596-7.\u003c/li\u003e\n\u003cli\u003eB\u0026ouml;sel J, Schiller P, Hook Y, Andes M, Neumann J-O, Poli S, et al. Stroke-related early tracheostomy versus prolonged orotracheal intubation in neurocritical care trial (SETPOINT): a randomized pilot trial. Stroke. 2013;44:21\u0026ndash;8. https://doi.org/10.1161/STROKEAHA.112.669895.\u003c/li\u003e\n\u003cli\u003eRobba C, Galimberti S, Graziano F, Wiegers EJA, Lingsma HF, Iaquaniello C, et al. Tracheostomy practice and timing in traumatic brain-injured patients: a CENTER-TBI study. Intensive Care Med. 2020;46:983\u0026ndash;94. https://doi.org/10.1007/s00134-020-05935-5.\u003c/li\u003e\n\u003cli\u003eSch\u0026ouml;nenberger S, Al-Suwaidan F, Kieser M, Uhlmann L, B\u0026ouml;sel J. The SETscore to predict tracheostomy need in cerebrovascular neurocritical care patients. Neurocrit Care. 2016;25:94\u0026ndash;104. https://doi.org/10.1007/s12028-015-0235-5.\u003c/li\u003e\n\u003cli\u003eChen X-H, Zhao J-J, Chen C, Yao L. Establishment and validation of a predictive model for tracheotomy in critically ill patients and analysis of the impact of different tracheotomy timing on patient prognosis. BMC Anesthesiol. 2024;24:175\u0026ndash;87. https://doi.org/10.1186/s12871-024-02558-x.\u003c/li\u003e\n\u003cli\u003eParry-Jones AR, Abid KA, Di Napoli M, Smith CJ, Vail A, Patel HC, et al. Accuracy and clinical usefulness of intracerebral hemorrhage grading scores: a direct comparison in a UK population. Stroke. 2013;44:1840\u0026ndash;5. https://doi.org/10.1161/STROKEAHA.113.001009.\u003c/li\u003e\n\u003cli\u003eDai J, Li S, Li X, Xiong W, Qiu Y. The mechanism of pathological changes of intraventricular hemorrhage in dogs. Neurol India. 2009;57:567\u0026ndash;77. https://doi.org/10.4103/0028-3886.57798.\u003c/li\u003e\n\u003cli\u003eJacobs B, Beems T, van der Vliet TM, Diaz-Arrastia RR, Borm GF, Vos PE. Computed tomography and outcome in moderate and severe traumatic brain injury: hematoma volume and midline shift revisited. J Neurotrauma. 2011;28:203\u0026ndash;15. https://doi.org/10.1089/neu.2010.1558.\u003c/li\u003e\n\u003cli\u003eDi Napoli M, Behrouz R, Topel CH, Misra V, Pomero F, Giraudo A, et al. Hypoalbuminemia, systemic inflammatory response syndrome, and functional outcome in intracerebral hemorrhage. J Crit Care. 2017;41:247\u0026ndash;53. https://doi.org/10.1016/j.jcrc.2017.06.002.\u003c/li\u003e\n\u003cli\u003eDuskin J, Yechoor N, Singh S, Mora S, Senff J, Kourkoulis C, et al. Nutrition markers and discharge outcome in deep and lobar intracerebral hemorrhage. Eur Stroke J. 2024;9:1074\u0026ndash;82. https://doi.org/10.1177/23969873241253048.\u003c/li\u003e\n\u003cli\u003eMorotti A, Boulouis G, Dowlatshahi D, Li Q, Shamy M, Al-Shahi Salman R, et al. Intracerebral haemorrhage expansion: definitions, predictors, and prevention. Lancet, Neurol. 2023;22:159\u0026ndash;71. https://doi.org/10.1016/S1474-4422(22)00338-6.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Table 1","content":"\u003cp\u003e\u003cstrong\u003eTable 1 The Study Population Demography\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"412\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTracheotomy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNon-TT (N=352)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTT (N=83)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eP-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAge (year)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e71.61 \u0026plusmn; 5.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e70.72 \u0026plusmn; 5.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.162\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTemperature (℃\u003c/strong\u003e\u003cstrong\u003e）\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e36.78 \u0026plusmn; 0.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e36.36 \u0026plusmn; 3.68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.749\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRespiratory rate (times/minute)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e18.94 \u0026plusmn; 4.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e18.10 \u0026plusmn; 6.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.219\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHeart rate (times/minute)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e83.67 \u0026plusmn; 15.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e83.73 \u0026plusmn; 17.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.53\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSystolic blood pressure (mmHg)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e160.49 \u0026plusmn; 24.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e168.22 \u0026plusmn; 28.26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDiastolic blood pressure (mmHg)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e88.29 \u0026plusmn; 15.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e91.18 \u0026plusmn; 16.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.069\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGlucose (mmol/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e7.86 \u0026plusmn; 2.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e8.24 \u0026plusmn; 2.46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.04\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eUric acid (umol/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e280.21 \u0026plusmn; 120.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e283.82 \u0026plusmn; 105.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.718\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePotassium (mmol/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e3.67 \u0026plusmn; 0.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e3.64 \u0026plusmn; 0.47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.643\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSodium (mmol/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e138.39 \u0026plusmn; 4.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e139.03 \u0026plusmn; 4.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.156\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCalcium (mmol/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e2.28 \u0026plusmn; 0.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e2.24 \u0026plusmn; 0.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.104\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePhosphorus \u0026nbsp;(mmol/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e0.91 \u0026plusmn; 0.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e0.91 \u0026plusmn; 0.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.629\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMagnesium \u0026nbsp;(mmol/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e0.86 \u0026plusmn; 0.13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e0.86 \u0026plusmn; 0.11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.449\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTotal protein (g/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e70.16 \u0026plusmn; 7.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e69.79 \u0026plusmn; 9.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.595\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAlbumin (g/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e38.95 \u0026plusmn; 4.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e40.09 \u0026plusmn; 5.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.025\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eWhite blood cell (10\u003csup\u003e9\u003c/sup\u003e/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e9.89 \u0026plusmn; 4.80\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e10.93 \u0026plusmn; 4.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNeutrophil (10\u003csup\u003e9\u003c/sup\u003e/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e8.07 \u0026plusmn; 4.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e9.37 \u0026plusmn; 4.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLymphocyte (10\u003csup\u003e9\u003c/sup\u003e/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e1.19 \u0026plusmn; 0.68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e1.14 \u0026plusmn; 0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.047\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMonocyte count (10\u003csup\u003e9\u003c/sup\u003e/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e0.51 \u0026plusmn; 0.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e0.51 \u0026plusmn; 0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.772\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHemoglobin (g/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e141.87 \u0026plusmn; 21.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e140.76 \u0026plusmn; 25.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.479\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePlatelet (10\u003csup\u003e9\u003c/sup\u003e/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e189.68 \u0026plusmn; 88.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e196.73 \u0026plusmn; 76.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.337\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eProthrombin time (s)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e12.65 \u0026plusmn; 6.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e11.63 \u0026plusmn; 0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.527\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eInternational normalized ratio\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e1.06 \u0026plusmn; 0.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e0.99 \u0026plusmn; 0.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.074\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eActivated partial thromboplastin time (s)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e26.33 \u0026plusmn; 6.45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e26.50 \u0026plusmn; 6.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.725\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eFibrinogen (g/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e3.20 \u0026plusmn; 2.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e3.15 \u0026plusmn; 1.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.346\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eThrombin time (s)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e16.79 \u0026plusmn; 2.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e16.87 \u0026plusmn; 2.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.832\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003col\u003e\n \u003cli\u003e\u003cstrong\u003edimer (mg/L)\u003c/strong\u003e\u003c/li\u003e\n \u003c/ol\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e2.61 \u0026plusmn; 11.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e1.78 \u0026plusmn; 3.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.017\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMidline shift (mm)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e1.85 \u0026plusmn; 3.36\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e4.34 \u0026plusmn; 4.39\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eInitial hematoma volume (mL)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e23.80 \u0026plusmn; 22.45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e36.89 \u0026plusmn; 20.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGCS\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e11.06 \u0026plusmn; 3.34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e8.16 \u0026plusmn; 2.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGender, (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.548\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eFemale\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e157 (44.60%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e34 (40.96%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMale\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e195 (55.40%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e49 (59.04%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHypertension, (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.025\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNo\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e102 (28.98%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e14 (16.87%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eYes\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e250 (71.02%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e69 (83.13%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDiabetes mellitus, (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.056\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNo\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e298 (84.66%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e63 (75.90%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eYes\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e54 (15.34%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e20 (24.10%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSmoking, (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.13\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNo\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e296 (84.09%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e64 (77.11%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eYes\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e56 (15.91%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e19 (22.89%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAlcohol consumption, (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.159\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNo\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e316 (89.77%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e70 (84.34%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eYes\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e36 (10.23%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e13 (15.66%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAnticoagulants, (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.297\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNo\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e318 (90.34%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e78 (93.98%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eYes\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e34 (9.66%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e5 (6.02%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eIntraventricular hemorrhage, (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNo\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e242 (68.75%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e34 (40.96%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eUnilateral\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e89 (25.28%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e40 (48.19%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBilateral\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e21 (5.97%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e9 (10.84%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSide of hematoma, (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.474\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLeft\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e167 (47.44%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e43 (51.81%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRight\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e185 (52.56%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e40 (48.19%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLesion location, (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBasal Ganglia\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e162 (46.02%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e58 (69.88%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLobar\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e79 (22.44%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e6 (7.23%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCerebellum\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e33 (9.38%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e6 (7.23%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBrainstem\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e6 (1.70%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e2 (2.41%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOther\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e26 (7.39%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e2 (2.41%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eThalamus\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e46 (13.07%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e9 (10.84%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCT signs, (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNo\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e269 (76.42%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e43 (51.81%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eIsland Sign\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e23 (6.53%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e9 (10.84%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBiphasic Sign\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e54 (15.34%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e31 (37.35%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eFluid Leve\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e4 (1.14%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e0 (0.00%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 32.767%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBlack Hole Sign\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.699%;\"\u003e\n \u003cp\u003e2 (0.57%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 24.0291%;\"\u003e\n \u003cp\u003e0 (0.00%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.5049%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eTT: Tracheotomy, GCS: Glasgow Coma Scale, CT: Computed Tomography\u003c/strong\u003e\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Intracerebral hemorrhage, Tracheotomy, Elderly, Machine learning, Random Forest, SHAP, Prediction model, Airway management","lastPublishedDoi":"10.21203/rs.3.rs-8177019/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8177019/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eRespiratory complications are common among elderly patients with intracerebral hemorrhage (ICH), which can cause more hypoxemia and neuronal injury, further leading to poor prognosis. Tracheotomy (TT) can effectively improve airway protection and oxygenation, but the timing of tracheotomy is usually determined by the experience of the attending clinician and lacks objective and standardized decision criteria. A reliable prediction tool for the early identification of elderly ICH patients who need TT is urgently needed. This study aimed to establish and validate a machine learning (ML) model to predict the early application of tracheotomy in elderly ICH patients.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eA total of 435 elderly patients (\u0026ge;\u0026thinsp;65 years) with ICH from two tertiary hospitals between 2015 and 2023 were retrospectively reviewed. A total of 40 clinical, laboratory, and imaging features were collected at admission. We applied LASSO regression for feature selection. Six machine learning algorithms (Random Forest, XGBoost, SVM, Gradient Boosting, Logistic Regression, and KNN) were trained and tested with a train\u0026ndash;test split of 70/30. The imbalanced class distribution was handled by SMOTE and class-weight adjustments. Hyperparameter optimization was conducted with Bayesian optimization. The performance of each algorithm was evaluated based on ROC\u0026ndash;AUC, PR curves, accuracy, precision, recall, and F1 score. SHAP values were calculated to interpret the models. The best-performing model was deployed as an online prediction tool.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eLASSO identified five predictors: Glasgow Coma Scale (GCS) score, intraventricular hemorrhage (IVH), midline shift, albumin level, and CT biphasic sign. Among the six algorithms, Random Forest performed the best (composite score 0.831; test AUC 0.723; accuracy 0.793; precision 0.784; recall 0.793; F1\u0026thinsp;=\u0026thinsp;0.788). SHAP analysis identified that GCS was the most important feature that contributed to tracheotomy risk, followed by IVH and midline shift. An online prediction tool was successfully deployed to estimate the early risk of tracheotomy in real time.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eA ML\u0026ndash;based model with five readily available clinical features accurately predicted the early need for TT in elderly ICH patients and may be helpful in timely airway management decisions.\u003c/p\u003e","manuscriptTitle":"Machine learning–based risk stratification for early tracheotomy in geriatric intracerebral hemorrhage: model development and web deployment","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-30 09:10:30","doi":"10.21203/rs.3.rs-8177019/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"fac9f890-2806-4bac-8a44-30257a69a7f9","owner":[],"postedDate":"December 30th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-04-29T07:55:49+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-30 09:10:30","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8177019","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8177019","identity":"rs-8177019","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}