A system for predicting delirium risk in a patient

Abstract

The application relates to a system for predicting delirium of a patient, the system comprising a data acquisition module for acquiring age, fever data, chest pain data, diarrhea data, vomiting data, dementia history data, coronary heart disease history data, renal insufficiency history data, tumor history data, blood oxygen saturation, GCS score, MEWS score, midazolam treatment data, deep vein puncture treatment data, nasal catheter oxygen inhalation treatment data, carbapenem antibiotic treatment data of the patient; and a module for calculating the delirium risk of the patient, which is used for calculating the data acquired in the data acquisition module, so as to calculate the delirium risk probability of the patient.

Description

Technical Field

[0001] This application belongs to the field of emergency medicine diagnostics, specifically relating to a system for predicting the risk of delirium in critically ill emergency patients. Background Technology

[0002] Delirium is an acute cognitive impairment syndrome characterized by altered levels of attention and consciousness, accompanied by multifaceted cognitive impairment. These symptoms typically develop over hours to days and exhibit fluctuating patterns. According to the latest definition in DSM-5-TR (2022), the diagnosis of delirium requires meeting the following criteria: ① Presence of attentional impairment, manifested as a decreased ability to concentrate, maintain, or divert attention, accompanied by altered levels of consciousness, i.e., reduced awareness of the environment; ② These impairments occur within a short period, typically from hours to days, and tend to fluctuate, with varying severity throughout the day; ③ Accompanied by other cognitive impairments, including memory deficits, orientation disorders, language impairments, visuospatial perception impairments, or perceptual disturbances; ④ These cognitive changes cannot be explained by pre-existing, established, or developing neurocognitive impairments, nor do they occur in a state of severe decreased consciousness (such as coma). Delirium can be classified into three subtypes: hyperactive, hypoactive, and mixed. Hyperactive delirium is characterized by marked agitation and hyperactivity, with patients typically exhibiting aggressive behavior, restlessness, and hallucinations / delusions. These patients are easily observed and identified due to their prominent symptoms, but are frequently misdiagnosed as having an acute episode of mental illness. Hypoactive delirium, on the other hand, is characterized by somnolence, slowed reflexes, reduced speech, and even decreased voluntary movement. Its presentation is often mistaken for depression or dementia and can be masked by other illnesses, leading to missed diagnoses. Mixed delirium involves alternating symptoms of both hyperactivity and hypoactivity, resulting in a more complex and variable clinical profile.

[0003] Critically ill emergency patients, as a special group of emergency patients, are typically adult patients who are directly admitted to the emergency resuscitation room or emergency intensive care unit for treatment after emergency triage. These patients usually have the following characteristics: ① They have one or more life-threatening clinical conditions, such as respiratory distress, hemodynamic instability, altered consciousness, or multiple organ dysfunction; ② They require high-intensity medical intervention, including but not limited to mechanical ventilation support, administration of vasoactive drugs, continuous organ function monitoring, or preparation for emergency surgery; ③ Their condition progresses rapidly and may deteriorate within a short period of time, requiring close observation and timely adjustment of treatment strategies. This definition is based on the allocation of medical resources and the assessment of patient severity in clinical practice, reflecting the special risks and needs faced by these patients in the emergency environment.

[0004] Currently, clinical risk assessment for delirium in critically ill patients mainly involves two technical approaches: one is based on bedside assessment scales, and the other is based on predictive models constructed from clinical variables. Regarding assessment scales, commonly used methods include consciousness assessment tools developed for the intensive care unit (ICU), such as the Confusion Assessment Method for Intensive Care Unit (CAM-ICU). These methods assess the patient's level of consciousness and cognitive function manually to determine the risk of delirium; however, these methods rely on manual operation and suffer from high subjectivity and insufficient real-time accuracy. Regarding predictive models, the closest existing technology to this application is a delirium prediction model based on the ICU population, including the PRE-DELIRIC model and its improved version, the E-PRE-DELIRIC model. The PRE-DELIRIC model integrates variables such as age, APACHE II score, metabolic acidosis, and medication use, and uses logistic regression to predict the risk of delirium. The E-PRE-DELIRIC model simplifies the variables to achieve earlier risk assessment. However, the above-mentioned existing technologies are mainly designed for ICU environments and have the following shortcomings in emergency scenarios: (1) They are highly dependent on data and cannot meet the real-time application needs of emergency departments: The above models rely on Apache. II score and multiple laboratory indicators, but such data are often not yet available or are incomplete in the early stage of emergency, which makes it impossible for the model to make effective predictions in the early stage of patient admission; (2) The variable system does not cover emergency-specific information: the existing model variables are mainly derived from ICU stable monitoring data, and do not include dynamic information generated during emergency treatment, such as chief complaint information, early intervention measures and medication information, thus failing to reflect key risk factors in the emergency scenario; (3) The model structure is not adaptable to complex data: the existing technology mainly uses linear statistical models, which have limited ability to model nonlinear relationships and interactions between variables. When clinical data is missing or unevenly distributed, the prediction stability is poor; (4) Lack of automated risk output mechanism: the existing methods mostly rely on manual assessment or independent calculation tools, and have not formed an automated process from data acquisition, model calculation to risk warning output, which is difficult to meet the needs of rapid decision-making in the emergency environment; (5) Time window lag: the existing models usually make predictions within a certain time window after admission to the ICU, and have not been optimized for the key time window in the early stage of emergency admission, affecting the timing of early intervention.

[0005] Existing delirium risk assessment tools (such as 4AT, CAM-ICU, and bCAM) face two key challenges in their application to critically ill emergency patients: First, the physiological parameters of critically ill emergency patients can fluctuate drastically within a short period, making it difficult for static scoring tools to capture such dynamic changes. Second, these patients often have multiple comorbidities and complex treatment plans, which limits the predictive accuracy of existing tools in this population and fails to meet the needs of early clinical identification and precise intervention. Therefore, there is an urgent clinical need for a tool that can rapidly and accurately predict the risk of delirium based on routine medical information, helping healthcare professionals to identify high-risk patients in advance and implement preventative interventions.

[0006] Machine learning methods offer new insights into solving these problems by integrating multidimensional data such as demographic characteristics, laboratory tests, vital signs, imaging data, and medication information. However, there is currently a lack of predictive and early warning models specifically designed for early delirium in critically ill emergency patients.

[0007] Existing delirium risk assessment techniques are primarily built upon intensive care unit (ICU) scenarios. Their model design relies on a relatively complete and stable clinical data environment and typically uses static severity scores (such as APACHE II) or fixed time windows for risk prediction. However, in the clinical application of these techniques to critically ill emergency patients, they have significant limitations and cannot meet the actual needs of rapid early assessment in the emergency department.

[0008] Specifically, the existing technology mainly has the following problems: (1) Difficulty in achieving real-time prediction in the early stage of emergency: Existing models rely on laboratory indicators and comprehensive scoring systems, but such data are often not yet available or are incomplete in the early stage of emergency admission, which makes it impossible for the model to conduct effective risk assessment within the critical time window; (2) Insufficient scenario adaptability: The existing prediction models are mainly based on ICU populations, and their variable system comes from a relatively stable monitoring environment. They have not been optimized for the characteristics of rapid changes in the condition of emergency patients and the dynamic complexity of the diagnosis and treatment process, which limits the applicability of the models in emergency scenarios. (3) Insufficient utilization of key clinical information: Existing technologies have failed to systematically incorporate important information that can be obtained in the early stages of the emergency department, such as chief complaint information, initial treatment measures and medication status, which means that some clinical information with predictive value has not been effectively utilized, thus affecting the accuracy of prediction results; (4) Limited ability to process complex data: Traditional prediction models based on linear statistical methods have limited ability to characterize the nonlinear relationships and interactions between variables when processing clinical data, and their stability and generalization ability are insufficient when faced with missing data and uneven data distribution. (5) Lack of automated risk assessment mechanism: Existing assessment methods mostly rely on manual scales or experience judgment, which have the problems of strong subjectivity and high operational dependence. It is difficult to achieve a standardized and automated risk prediction process and cannot meet the needs of rapid decision support in the emergency environment.

[0009] The aforementioned problems arise primarily from the fact that existing technologies are built upon the relatively stable data acquisition environment of the ICU. Their variable systems and model designs are not optimized for emergency treatment processes and are highly dependent on data integrity, thus limiting their application in the early stages of emergency care. Therefore, it is necessary to propose a delirium risk prediction method suitable for emergency scenarios. This method should be able to operate stably even with incomplete data and achieve early, objective, and automated assessment of patient delirium risk, overcoming the shortcomings of existing technologies such as strong data dependence, insufficient scenario adaptability, and high subjectivity in the assessment process. Summary of the Invention

[0010] This application addresses the clinical problem of predicting delirium in critically ill emergency patients by constructing a model through a multi-stage research design. The study systematically assesses the current status of delirium management in critically ill emergency patients through a nationwide multi-center cross-sectional survey, identifying key risk factors for clinicians' perceptions. Based on this, it integrates readily available variables in the emergency department environment, such as demographic characteristics, chief complaints, comorbidities, vital signs, and early laboratory tests, and utilizes machine learning techniques to construct an early prediction and warning model for delirium in critically ill emergency patients. External validation using multi-center data evaluates the model's generalization ability and clinical applicability, providing an evidence-based tool for the early identification and intervention of delirium in critically ill emergency patients.

[0011] Specifically, this application involves the following: 1. A system for predicting the risk of delirium in patients, comprising: The data acquisition module is used to acquire patient data including age, fever, chest pain, diarrhea, vomiting, history of dementia, history of coronary heart disease, history of renal insufficiency, history of cancer, blood oxygen saturation, Glasgow Coma Scale (GCS) score, Meniere's Disease Scale (MEWS) score, midazolam treatment data, deep vein puncture treatment data, nasal cannula oxygen therapy treatment data, and carbapenem antibiotic treatment data; as well as... The module for calculating the risk of delirium in patients is used to perform calculations on the data acquired by the data acquisition module to determine the probability of delirium in patients. .

[0012] 2. The system according to item 1, wherein the age, blood oxygen saturation, GCS score, and MEWS score are continuous variables.

[0013] 3. According to the system described in item 1, wherein the fever data, chest pain data, diarrhea data, vomiting data, history of dementia data, history of coronary heart disease data, history of renal insufficiency data, history of tumor data, midazolam treatment data, deep vein puncture treatment data, nasal cannula oxygen therapy data, and carbapenem antibiotic treatment data are binary variables.

[0014] 4. The system according to item 3, wherein, If the patient's body temperature is greater than or equal to 37.3°C, the fever data is 1; if the patient's body temperature is less than 37.3°C, the fever data is 0. The patient has chest pain symptoms, and the chest pain data is 1; the patient has no chest pain symptoms, and the chest pain data is 0. If a patient has diarrhea symptoms, the diarrhea data is 1; if a patient does not have diarrhea symptoms, the diarrhea data is 0. If the patient has vomiting symptoms, the vomiting data is 1; if the patient does not have vomiting symptoms, the vomiting data is 0. The patient has a history of dementia, and the data for the history of dementia is 1; the patient has no history of dementia, and the data for the history of dementia is 0. The patient has a history of coronary heart disease, and the data for the history of coronary heart disease is 1; the patient has no history of coronary heart disease, and the data for the history of coronary heart disease is 0. The patient has a history of renal insufficiency, and the data for the history of renal insufficiency is 1; the patient has no history of renal insufficiency, and the data for the history of renal insufficiency is 0. The patient has a history of cancer, and the cancer history data is 1; the patient has no history of cancer, and the cancer history data is 0. The patient is being treated with midazolam, and the midazolam treatment data is 1; the patient is not being treated with midazolam, and the midazolam treatment data is 0. The patient has undergone deep vein puncture treatment, and the deep vein puncture treatment data is 1; the patient has not undergone deep vein puncture treatment, and the deep vein puncture treatment data is 0. The patient received nasal cannula oxygen therapy, and the nasal cannula oxygen therapy data is 1; the patient did not receive nasal cannula oxygen therapy, and the nasal cannula oxygen therapy data is 0. The patient received carbapenem antibiotic treatment, and the carbapenem antibiotic treatment data is 1; the patient did not receive carbapenem antibiotic treatment, and the carbapenem antibiotic treatment data is 0.

[0015] 5. The system according to items 1-4, wherein the module for calculating the risk of delirium in a patient pre-stores data based on age, fever, chest pain, diarrhea, vomiting, history of dementia, history of coronary heart disease, history of renal insufficiency, history of cancer, blood oxygen saturation, Glasgow Coma Scale (GCS) score, MEWS score, midazolam treatment data, deep vein puncture treatment data, nasal cannula oxygen therapy data, and carbapenem antibiotic treatment data, fitted by logistic regression to calculate the probability of delirium risk in a patient. The formula.

[0016] 6. The system according to item 5, wherein the formula is: in, These are the initial parameters. For the patient's age, Its corresponding regression coefficient; For fever data, Its corresponding regression coefficient; Data for chest pain; Its corresponding regression coefficient; Data for diarrhea; Its corresponding regression coefficient; Data on vomiting; Its corresponding regression coefficient; For dementia history data, Its corresponding regression coefficient; For data on past history of coronary heart disease, Its corresponding regression coefficient; Data on a history of renal insufficiency Its corresponding regression coefficient; For data on past cancer history, Its corresponding regression coefficient; Blood oxygen saturation Its corresponding regression coefficient; GCS score, Its corresponding regression coefficient; Rate MEWS Its corresponding regression coefficient; Data on midazolam treatment. Its corresponding regression coefficient; Data for deep vein puncture treatment. Its corresponding regression coefficient; Data on nasal cannula oxygen therapy. Its corresponding regression coefficient; Data on carbapenem antibiotic treatment. Its corresponding regression coefficient.

[0017] 7. The system according to item 6, wherein, It can be any value selected from 0.022 to 0.032; It can be any value selected from -0.686 to -0.458; For any value selected from -0.584 to -0.390; It can be any value selected from -1.256 to -0.838; It can be any value selected from -0.840 to -0.560; It can be any value selected from 0.210 to 0.316; It can be any value selected from -1.452 to -0.968; It can be any value selected from 1.036 to 1.554; It can be any value selected from 1.497 to 2.245; It can be any value selected from -0.043 to -0.029; It can be any value selected from -0.518 to -0.346; It can be any value selected from 0.123 to 0.185; It can be any value selected from 1.654 to 2.482; It can be any value selected from 0.654 to 0.980; It can be any value selected from 0.694 to 1.040; It can be any value selected from 0.710 to 1.064.

[0018] 8. The system according to item 7, wherein, The preferred value is 0.027; The preferred value is -0.572; The preferred value is -0.487; The preferred value is -1.047; The preferred value is -0.700; The preferred value is 0.263; The preferred value is -1.21; The preferred value is 1.295; The preferred value is 1.871; The preferred value is -0.036; The preferred value is -0.432; The preferred value is 0.154; The preferred value is 2.068; The preferred value is 0.817; The preferred value is 0.867; The preferred value is 0.887.

[0019] 9. The system according to item 1, further comprising: The grouping module contains pre-stored default delirium risk grouping parameters, and based on these parameters, it assigns the calculated probability of delirium risk to the subject. Subjects were grouped to assess their risk of delirium.

[0020] 10. A method for predicting the risk of delirium in a patient, comprising: The data collection steps are used to obtain patient data on age, fever, chest pain, diarrhea, vomiting, history of dementia, history of coronary heart disease, history of renal insufficiency, history of cancer, blood oxygen saturation, Glasgow Coma Scale (GCS) score, Meniere's Disease Scale (MEWS) score, midazolam treatment data, deep vein puncture treatment data, nasal cannula oxygen therapy treatment data, and carbapenem antibiotic treatment data; and The step of calculating the risk of delirium in a patient involves processing the data acquired from the data acquisition module to determine the probability of delirium. .

[0021] 11. The method according to item 10, wherein the age, blood oxygen saturation, GCS score, and MEWS score are continuous variables.

[0022] 12. The method according to item 10, wherein the fever data, chest pain data, diarrhea data, vomiting data, history of dementia data, history of coronary heart disease data, history of renal insufficiency data, history of tumor data, midazolam treatment data, deep vein puncture treatment data, nasal cannula oxygen therapy data, and carbapenem antibiotic treatment data are binary variables.

[0023] 13. The method according to item 12, wherein, If the patient's body temperature is greater than or equal to 37.3°C, the fever data is 1; if it is less than 37.3°C, the fever data is 0. The patient has chest pain symptoms, and the chest pain data is 1; the patient has no chest pain symptoms, and the chest pain data is 0. If a patient has diarrhea symptoms, the diarrhea data is 1; if a patient does not have diarrhea symptoms, the diarrhea data is 0. If the patient has vomiting symptoms, the vomiting data is 1; if the patient does not have vomiting symptoms, the vomiting data is 0. The patient has a history of dementia, and the data for the history of dementia is 1; the patient has no history of dementia, and the data for the history of dementia is 0. The patient has a history of coronary heart disease, and the data for the history of coronary heart disease is 1; the patient has no history of coronary heart disease, and the data for the history of coronary heart disease is 0. The patient has a history of renal insufficiency, and the data for the history of renal insufficiency is 1; the patient has no history of renal insufficiency, and the data for the history of renal insufficiency is 0. The patient has a history of cancer, and the cancer history data is 1; the patient has no history of cancer, and the cancer history data is 0. The patient is being treated with midazolam, and the midazolam treatment data is 1; the patient is not being treated with midazolam, and the midazolam treatment data is 0. The patient has undergone deep vein puncture treatment, and the deep vein puncture treatment data is 1; the patient has not undergone deep vein puncture treatment, and the deep vein puncture treatment data is 0. The patient received nasal cannula oxygen therapy, and the nasal cannula oxygen therapy data is 1; the patient did not receive nasal cannula oxygen therapy, and the nasal cannula oxygen therapy data is 0. The patient received carbapenem antibiotic treatment, and the carbapenem antibiotic treatment data is 1; the patient did not receive carbapenem antibiotic treatment, and the carbapenem antibiotic treatment data is 0.

[0024] 14. The method according to items 10-13, wherein, in the step of calculating the risk of delirium in a patient, a probability of delirium risk for calculating the patient is pre-stored, which is fitted by logistic regression based on age, fever data, chest pain data, diarrhea data, vomiting data, history of dementia data, history of coronary heart disease data, history of renal insufficiency data, history of tumor data, blood oxygen saturation, GCS score, MEWS score, midazolam treatment data, deep vein puncture treatment data, nasal cannula oxygen therapy treatment data, and carbapenem antibiotic treatment data. The formula for ).

[0025] 15. The method according to item 14, wherein the formula is: in, These are the initial parameters. For the patient's age, Its corresponding regression coefficient; For fever data, Its corresponding regression coefficient; Data for chest pain; Its corresponding regression coefficient; Data for diarrhea; Its corresponding regression coefficient; Data on vomiting; Its corresponding regression coefficient; For dementia history data, Its corresponding regression coefficient; For data on past history of coronary heart disease, Its corresponding regression coefficient; Data on a history of renal insufficiency Its corresponding regression coefficient; For data on past cancer history, Its corresponding regression coefficient; Blood oxygen saturation Its corresponding regression coefficient; GCS score, Its corresponding regression coefficient; Rate MEWS Its corresponding regression coefficient; Data on midazolam treatment. Its corresponding regression coefficient; Data for deep vein puncture treatment. Its corresponding regression coefficient; Data on nasal cannula oxygen therapy. Its corresponding regression coefficient; Data on carbapenem antibiotic treatment. Its corresponding regression coefficient.

[0026] 16. The method according to item 15, wherein, It can be any value selected from 0.022 to 0.032; It can be any value selected from -0.686 to -0.458; For any value selected from -0.584 to -0.390; It can be any value selected from -1.256 to -0.838; It can be any value selected from -0.840 to -0.560; It can be any value selected from 0.210 to 0.316; It can be any value selected from -1.452 to -0.968; It can be any value selected from 1.036 to 1.554; It can be any value selected from 1.497 to 2.245; It can be any value selected from -0.043 to -0.029; It can be any value selected from -0.518 to -0.346; It can be any value selected from 0.123 to 0.185; It can be any value selected from 1.654 to 2.482; It can be any value selected from 0.654 to 0.980; It can be any value selected from 0.694 to 1.040; It can be any value selected from 0.710 to 1.064.

[0027] 17. The method according to item 16, wherein, The preferred value is 0.027; The preferred value is -0.572; The preferred value is -0.487; The preferred value is -1.047; The preferred value is -0.700; The preferred value is 0.263; The preferred value is -1.21; The preferred value is 1.295; The preferred value is 1.871; The preferred value is -0.036; The preferred value is -0.432; The preferred value is 0.154; The preferred value is 2.068; The preferred value is 0.817; The preferred value is 0.867; The preferred value is 0.887.

[0028] 18. The method according to item 10, further comprising: The grouping step includes a pre-stored default delirium risk grouping parameter, and based on this parameter, the probability of delirium risk in the calculated subject is determined. Subjects were grouped to assess their risk of delirium.

[0029] Invention Effects This application addresses the shortcomings of existing technologies in the early prediction of delirium in critically ill emergency patients, such as insufficient predictive timeliness, incomplete variable system, and poor model adaptability. By improving the variable construction method, data processing strategy, and model implementation method, the following technical effects have been achieved: In terms of predictive performance, this application, based on traditional physiological indicators, introduces early emergency patient complaints and treatment intervention factors to construct a predictive variable system that better reflects the temporal characteristics of emergency situations. Through these variable enhancements, the model can more accurately characterize the patient's pathological state in the early stages of admission, thereby significantly improving predictive ability. The predictive model built based on this variable system achieves an area under the receiver operating characteristic (AUC) of 0.899, demonstrating higher predictive accuracy and earlier risk identification capabilities in emergency scenarios compared to existing predictive models primarily based on intensive care environments.

[0030] Regarding data adaptability, this application addresses missing data by introducing a predictive mean matching method and incorporating robust design for incomplete data into the model. This enables the prediction method to adapt to the high missing data rate and significant data quality fluctuations characteristic of emergency environments. Even with missing key variables, it maintains stable predictive performance, thereby improving the model's applicability in real-world clinical settings.

[0031] In clinical applications, this application constructs an automated risk assessment process that can complete risk calculation and grading output in real time after obtaining basic patient information and early diagnosis and treatment data. This method can achieve early identification of high-risk patients without requiring additional steps from medical staff, thereby reducing reliance on manual assessment and improving the objectivity and responsiveness of clinical decision-making.

[0032] Furthermore, this application has low requirements for data integrity and equipment conditions, and can be implemented in emergency environments of medical institutions at different levels, demonstrating good scalability. By identifying high-risk patients in advance and optimizing the timing of intervention, the incidence of delirium and related medical costs can be reduced to a certain extent, demonstrating significant clinical application value and economic benefits. Attached Figure Description

[0033] Various other advantages and benefits of this application will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiments below. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0034] Figure 1 The diagram shows the coefficient variation trajectory of 78 candidate variables screened by LASSO regression under different regularization parameters according to an embodiment of this application.

[0035] Figure 2 The graph shows the relationship between the 10-fold cross-validation error (CVM) of LASSO regression and Log(λ) in an embodiment of this application.

[0036] Figure 3 This paper presents a nomogram for predicting delirium in critically ill emergency patients based on a logistic regression model, as described in an embodiment of this application.

[0037] Figure 4 The diagram shows a comparison of the ROC curves of the logistic regression model of this application and the E-PRE-DELIRIC model on a random single-split validation set.

[0038] Figure 5 The diagram shows the receiver operating characteristic (ROC) curves of the logistic regression model and three other prediction models described in this application on an independent external validation set of the emergency critical patient cohort of Peking University First Hospital.

[0039] Figure 6 The diagram shows a comparison between the logistic regression model described in this application and the E-PRE-DELIRIC scoring system on an independent external validation set of critically ill patients in the emergency department of Peking University First Hospital.

[0040] Figure 7 The diagram shows the receiver operating characteristic (ROC) curves of the logistic regression model described in this application and three other prediction models on the MIMIC-IV database validation set.

[0041] Figure 8 Calibration curves showing the consistency between the predicted probabilities of the model described in this application and three other machine learning models on the MIMIC-IV data validation set and the actual observations are presented.

[0042] Figure 9The user interface of the delirium risk prediction system for critically ill emergency patients according to an embodiment of this application is shown. Detailed Implementation

[0043] Specific embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While specific embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0044] It should be noted that certain terms are used in the specification and claims to refer to specific components. Those skilled in the art will understand that different terms may be used to refer to the same component. This specification and claims do not distinguish components based on differences in terminology, but rather on differences in function. The terms "comprising" or "including" used throughout the specification and claims are open-ended and should be interpreted as "comprising but not limited to." The following descriptions in the specification are preferred embodiments for carrying out this application; however, these descriptions are for the purpose of understanding the general principles of the specification and are not intended to limit the scope of this application. The scope of protection of this application shall be determined by the appended claims.

[0045] Variable types: In statistics, variable types can be divided into two types: quantitative variables and qualitative variables (also known as categorical variables).

[0046] Quantitative variables are used to describe the quantity and number of things, and can be further divided into continuous and discrete variables. Continuous variables are variables that can take any value within a certain range; their values ​​are continuous and can have decimal points. Examples include blood pressure, blood sugar, and anthropometric measurements such as height, weight, and chest circumference. Their values ​​can only be obtained through measurement or quantitative methods. Discrete variables are variables whose values ​​can only be natural numbers or integer units. Examples include pain scores, the number of metastatic lesions, and the number of retrieved oocytes; these can only be positive numbers and cannot have decimal points. The values ​​of such variables are generally obtained using counting methods.

[0047] Logistic regression is a generalized linear regression analysis model commonly used in data mining, automated disease diagnosis, and economic forecasting. For example, it can be used to explore risk factors for diseases and predict the probability of disease occurrence based on these risk factors. Taking gastric cancer analysis as an example, two groups are selected: one with gastric cancer and the other without. These two groups will inevitably have different physical characteristics and lifestyles. Therefore, the dependent variable is whether or not someone has gastric cancer, with a value of "yes" or "no." The independent variables can include many factors, such as age, gender, dietary habits, and Helicobacter pylori infection. The independent variables can be continuous or categorical. Through logistic regression analysis, the weights of the independent variables can be obtained, allowing for a general understanding of which factors are risk factors for gastric cancer. Furthermore, based on these weights, the likelihood of an individual developing cancer can be predicted based on these risk factors. The dependent variable in logistic regression can be binary or multi-category.

[0048] The data used in this paper are fitted to a logistic regression model, which penalizes the absolute magnitude of the coefficients of the regression model based on the value of λ. The larger the penalty, the closer the estimates of weaker factors approach zero, so only the strongest predictor variables are retained in the model.

[0049] The term "critically ill emergency patient" as used in this application refers to an adult patient who is directly admitted to the emergency resuscitation room or emergency intensive care unit after emergency triage and requires emergency treatment and continuous monitoring due to the severity of their condition.

[0050] The model in this application aims to predict whether critically ill emergency room patients will develop delirium (CAM-ICU as the dependent variable). Since this is a binary classification problem, we use a confusion matrix to evaluate model performance. The confusion matrix contains four basic counts: True Positive (TP), which is when the model correctly predicts a patient will develop delirium; False Negative (FN), which is when the model incorrectly predicts a patient will not develop delirium but actually does; False Positive (FP), which is when the model incorrectly predicts a patient will develop delirium but actually does not; and True Negative (TN), which is when the model correctly predicts a patient will not develop delirium. These counts provide a comprehensive understanding of the model's accuracy and reliability in predicting the occurrence of delirium.

[0051] (1) Accuracy: The proportion of correct predictions out of all predictions. The calculation formula is as follows: Accuracy = (TP + TN) / (TP + TN + FP + FN) (2) Sensitivity: the proportion of actual positive samples that are correctly predicted as positive. The calculation formula is as follows: Sensitivity = TP / (TP + FN) (3) Specificity: the proportion of actual negative samples that are correctly predicted as negative. The calculation formula is as follows: Specificity = TN / (TN + FP) (4) Precision: the proportion of true positives among the samples predicted as positive. The calculation formula is as follows: Precision = TP / (TP + FP) (5) Youden Index: a comprehensive indicator of sensitivity and specificity. The calculation formula is as follows: Youden Index = Sensitivity + Specificity – 1 (6) Receiver Operating Characteristic curve (ROC curve) and Area Under the Curve (AUC): The ROC curve evaluates the classification performance of a model by plotting sensitivity (true positive rate) against (1 - Specificity) (false positive rate) at different decision thresholds. AUC is a quantitative indicator of the model's discrimination ability. The evaluation criteria for AUC values are: 0.5 < AUC < 0.7 indicates low accuracy; 0.7 ≤ AUC < 0.9 indicates moderate accuracy; AUC ≥ 0.9 indicates high accuracy. The DeLong method in the pROC package is used for the comparison of areas under the curves.

[0052] This application, based on a cross-sectional survey of 402 emergency physicians nationwide and a prospective cohort of 928 critically ill emergency patients, systematically analyzed the risk factors for delirium. The results showed a general lack of standardized monitoring tools in clinical practice, with physicians paying close attention to core risk factors such as metabolic disorders, pre-existing cognitive impairment, and shock. The prospective cohort observed a delirium incidence rate of 17.1% (159 / 928), consistent with previously reported ranges of 10%–30%. Advanced age was the most significant independent risk factor; the median age of patients in the delirium group was 78 years, significantly higher than the 67 years in the non-delirium group (p<0.001). Furthermore, certain underlying diseases also constituted important susceptibility factors in this cohort, including a history of cancer (5.7% vs 0.8%, p<0.001) and a history of renal insufficiency (10.1% vs 4.9%, p=0.020). This confirms the multifactorial cumulative view of delirium proposed by Inouye et al., that is, the higher the patient's baseline susceptibility (e.g., advanced age, multiple underlying diseases), the greater the risk of developing delirium. In terms of physical examination, patients in the delirium group exhibited more pronounced physiological disturbances upon admission. The median respiratory rate was slightly higher in the delirium group than in the non-delirium group (18 breaths / min vs 17 breaths / min, p=0.006), oxygen saturation was significantly lower (96% vs 98%, p=0.004), diastolic blood pressure was lower (67 mmHg vs 68 mmHg, p=0.034), Glasgow Coma Scale (GCS) score was significantly lower (14 vs 15, p<0.001), and MEWS score was significantly higher (3 vs 2, p<0.001). These abnormal physiological indicators reflect that delirium patients were more severely ill upon admission and may have underlying respiratory and circulatory dysfunction. Prolonged emergency observation time was closely related to the occurrence of delirium; the median observation time in the delirium group was 86.5 hours, significantly longer than the 39.0 hours in the non-delirium group (P<0.001), indicating that prolonged emergency observation time is a risk factor for delirium. This may be related to the inherent characteristics of the emergency room environment (such as noise, light, sleep deprivation, frequent interruptions, and lack of orientation cues) and the continued lack of effective control over the patient's condition. These environmental factors were also identified as risk factors for delirium by emergency room physicians in the investigation in Chapter 3. Laboratory analysis showed that patients in the delirium group had multiple abnormal biochemical indicators, reflecting a stronger inflammatory response and metabolic disturbances. The white blood cell count was significantly elevated (median 10.05 × 10⁻⁶). 9 / L vs 9.00×10 9The levels of lactate dehydrogenase (277 U / L vs 251 U / L, p = 0.002) were significantly lower than those of 251 U / L, suggesting a stronger inflammatory response or tissue damage. Serum albumin levels were also significantly lower (34.7 g / L vs 37.3 g / L, p < 0.001), reflecting malnutrition or a negative nitrogen balance, consistent with previous studies suggesting malnutrition as a risk factor for delirium. Elevated blood urea nitrogen levels (10.79 mmol / L vs 8.56 mmol / L, p < 0.001) further support the important role of metabolic disorders in the development of delirium. In a subset of patients (N=123) who underwent head imaging, this application found that the detection rate of age-related brain changes (brain atrophy, ventricular enlargement) was significantly higher in the delirium group (30.6% vs 12.6%, p=0.036), while the detection rate of white matter lesions was higher in the delirium group (58.3% vs 40.2%), but the difference was not statistically significant (p=0.102). This is consistent with the view of Cavalli et al. that changes in brain structure (especially white matter lesions and atrophy) increase susceptibility to delirium

[139] . However, it should be noted that there was a significant selection bias in the imaging examinations in this application, and the results should be interpreted with caution. Regarding treatment-related factors, the use of sedative drugs such as midazolam (P<0.001) and dexmedetomidine (P=0.017), as well as carbapenem antibiotics (P<0.001), was more common in the delirium group.

[0053] In addition, the proportions of nasal cannula oxygen therapy (P=0.002) and deep vein puncture (P<0.001) were also higher. The relationship between these factors and delirium is complex. On the one hand, sedative drugs (especially benzodiazepines, such as midazolam) are themselves considered risk factors for delirium; on the other hand, the use of these drugs and procedures also reflects the severity of the patient's condition or specific treatment needs (such as sedation before mechanical ventilation, treatment of severe infection). When discussing these factors as risk factors, the possibility of confounding and bidirectional associations should be considered

[138] . This application further validates the close association between delirium and poor prognosis. In the modeled cohort, the mortality rate of patients with delirium (16.35%) was significantly higher than that of the non-delirium group (5.20%, P<0.001), highlighting the importance of early identification and intervention of delirium. This application compares the differences in clinical characteristics between patients with high activity (56.0%) and low activity (41.5%) delirium. Patients with low-activity delirium are typically older, have a greater history of neurological disorders (e.g., stroke 22.7% vs 9.0%, Parkinson's disease 6.1% vs 0.0%), lower Glasgow Coma Scale (GCS) scores on admission, require a higher proportion of respiratory support (non-invasive / invasive ventilation), and have longer hospital stays. This suggests that low-activity delirium may be associated with more severe illness or a poorer underlying condition, and its symptoms are often subtle and easily overlooked. In contrast, patients with high-activity delirium are relatively younger, have higher GCS scores on admission, but may exhibit more pronounced vital sign abnormalities (e.g., elevated heart rate, elevated body temperature) and are more likely to receive antipsychotic medications or sedatives due to their behavioral symptoms. These differences suggest that different delirium subtypes may be associated with different risk factors and clinical trajectories, which is important for accurate assessment and management. Due to the limited sample size of the subtype analysis, these results mainly suggest associations and require further confirmation with larger sample applications. In summary, the cohort analysis presented in this application revealed multidimensional risk factors for delirium in critically ill emergency patients, encompassing patient baseline conditions (advanced age, specific underlying diseases), physiological indicators upon admission (abnormal vital signs, abnormal laboratory indicators), potential susceptibility to brain structures (age-related brain changes), and emergency environment and treatment-related factors (observation time, sedatives, antibiotics, central venous access, etc.). The identification of these risk factors not only lays the foundation for constructing accurate predictive models but also provides important clues for clinicians to develop early, multifaceted prevention and intervention strategies.

[0054] This application employs a multi-dimensional approach to identify and handle outliers: considering that most of the predictor variables in this application are categorical variables, we focused on verifying the logical relationships between variables. For example, we checked whether records of the use of a specific antibiotic (such as cephalosporins) were consistent with their corresponding categorical variable of "whether antibiotics were used". For data points that did not conform to the preset logical judgment, we traced and verified them by reviewing the original electronic medical records and then corrected them.

[0055] For handling missing values, the following strategy is adopted: (1) Missing pattern analysis: The missing value pattern was analyzed using the "naniar" package in R language. Little's MCAR test was used to assess the missing mechanism. At the same time, clinical expertise was combined to determine the possible missing mechanism type of the variable, including Missing Completely At Random (MCAR), Missing At Random (MAR), and Missing Not At Random (MNAR).

[0056] (2) Sample removal: Patients who are missing the main outcome variable (occurrence of delirium) or the key tool scores necessary for outcome assessment (such as the components of the CAM-ICU score) are directly removed from the dataset because they cannot be effectively analyzed.

[0057] (3) Variable removal: After removing the above samples, the missing rate of each predictor variable in the remaining dataset is calculated. We set a threshold of missing rate > 20%. Any variable exceeding this threshold is considered to have severely insufficient information and is difficult to impute reliably, so it is removed from the subsequent model building analysis.

[0058] (4) Multiple interpolation: Multiple interpolation is performed using the mice package in R language to generate m=10 complete datasets.

[0059] During model construction, models are first trained separately on each imputation dataset, and then the results are integrated according to the Rubin merging rule. For the logistic regression model, the regression coefficients are merged using the average method, and the standard error is calculated using the Rubin formula. For the machine learning model, an ensemble learning method is used, averaging the predictions of 10 models to obtain the final prediction result. For all remaining predictors in the dataset with a missing rate ≤20%, we use multiple imputation, employing Predictive Mean Matching (PMM) as the imputation algorithm.

[0060] Regarding parameter selection, the parameter selection in this application includes the following steps. (1) Delete variables that are not related to the patient’s condition (such as medical record number, time of admission to the emergency room, etc.).

[0061] (2) Delete variables with a missing percentage >20% (procalcitonin, D-dimer quantification, whole blood troponin I, etc.).

[0062] (3) Active exclusion of head imaging-related variables: This application collected three variables related to head imaging: whether a head imaging examination was performed (binary variable, 1=yes, 0=no), white matter lesions (binary variable, 1=yes, 0=no), and age-related brain changes (binary variable, 1=yes, 0=no). Although these examinations were all performed in the emergency room, in clinical practice, only a portion of patients (129 cases, accounting for 12.2%) underwent head imaging examinations as medically necessary, and most patients did not undergo such examinations. Considering the actual clinical application scenario, patients may not have decided whether they need head imaging examinations at the beginning of their hospitalization, and including these variables in the predictive model may limit the real-time application value of the model. In addition, the decision to undergo head imaging examinations may itself be related to the severity of the patient's symptoms, and there is a certain indication bias. Based on the above considerations, we actively excluded these three variables in the modeling process to construct a more universal and clinically applicable predictive model.

[0063] (4) Variable selection via LASSO regression: LASSO is a regression analysis method that can simultaneously perform variable selection and regularization. Its core idea is to penalize the regression coefficients in the model, thereby compressing the coefficients of some unimportant variables to zero, thus achieving variable selection. The purpose of LASSO regression is to find a balance between predictive accuracy and model simplicity. In this application, the glmnet package in R language is used for LASSO regression analysis. The optimal penalty parameter λ is determined by 10-fold cross-validation. This process is repeated 10 times on the training set, with 9 sets of data used to train the model and 1 set of data used for validation each time. Finally, the average cross-validation error of the model under different λ values ​​is calculated to obtain a model that has both good predictive performance and contains as few variables as possible. We evaluate the parameter value (λ_min) that produces the minimum error in the cross-validation process and the parameter value (λ_1se) corresponding to a more minimalist model within one standard error range.

[0064] This study used LASSO regression to select 16 variables with independent predictive value from 78 candidate variables, covering multiple clinical dimensions such as demographic characteristics, clinical symptoms, physical examination, and treatment interventions. This variable selection strategy effectively controlled model complexity while ensuring model predictive performance, thus guaranteeing clinical operability in emergency settings.

[0065] This application employs four machine learning algorithms to construct a predictive and early warning model for early delirium in critically ill emergency patients: logistic regression, extreme gradient boosting, classification boosting, and random forest. The performance of each method is compared. All algorithms are implemented in Python (version 3.10), with logistic regression and random forest using the scikit-learn library (version 1.4.1), XGBoost using the XGBoost library (version 1.5.0), and CatBoost using the CatBoost library (version 0.26.1). All models use a fixed random seed (random_state=1) to ensure reproducibility of results.

[0066] Finally, to address the problems existing in the prior art, this application proposes a system for predicting the risk of delirium in patients, comprising: a data acquisition module for acquiring patient data on age, fever, chest pain, diarrhea, vomiting, history of dementia, history of coronary heart disease, history of renal insufficiency, history of cancer, blood oxygen saturation, Glasgow Coma Scale (GCS) score, Meniere's Disease Scale (MEWS) score, midazolam treatment data, deep vein puncture treatment data, nasal cannula oxygen therapy data, and carbapenem antibiotic treatment data; and a module for calculating the risk of delirium, which calculates the probability of delirium risk in patients based on the data acquired in the data acquisition module. .

[0067] In this article, there are no restrictions on the data acquisition module, as long as it can be used to obtain the above-mentioned data from the patient.

[0068] In the system described in this application, age, blood oxygen saturation, Glasgow Coma Scale (GCS) score, and Meniere's Disease Score (MEWS) score are continuous variables. Fever data, chest pain data, diarrhea data, vomiting data, history of dementia data, history of coronary heart disease data, history of renal insufficiency data, history of cancer data, midazolam treatment data, central venous catheterization treatment data, nasal cannula oxygen therapy treatment data, and carbapenem antibiotic treatment data are binary variables. These variables integrate demographic characteristics, chief complaints, comorbidities, vital signs, and early laboratory tests—variables easily obtained in an emergency setting.

[0069] In one specific implementation, for continuous variables such as age, blood oxygen saturation, GCS score, and MEWS score, the system collects their actual values ​​as variable values.

[0070] In one specific implementation, for the binary classification variables collected by the system, values ​​are assigned as "Yes" = 1 and "No" = 0. For example: if a patient's body temperature is greater than or equal to 37.3℃, the fever data is 1; if a patient's body temperature is less than 37.3℃, the fever data is 0; if a patient has chest pain, the chest pain data is 1; if a patient does not have chest pain, the chest pain data is 0; if a patient has diarrhea, the diarrhea data is 1; if a patient does not have diarrhea, the diarrhea data is 0; if a patient has vomiting, the vomiting data is 1; if a patient does not have vomiting, the vomiting data is 0; if a patient has a history of dementia, the dementia history data is 1; if a patient does not have a history of dementia, the dementia history data is 0; if a patient has a history of coronary heart disease, the coronary heart disease history data is 1; if a patient does not have a history of coronary heart disease, the coronary heart disease history data is 0; if a patient has a history of renal insufficiency, the renal insufficiency history data is... The following data points are considered in relation to the patient's condition: 1 for a patient with no history of renal insufficiency (data for history of renal insufficiency is 0); 1 for a patient with a history of cancer (data for history of cancer is 1), 0 for a patient without a history of cancer (data for history of cancer); 1 for a patient treated with midazolam (data for midazolam treatment is 1), 0 for a patient who did not use midazolam (data for midazolam treatment is 0); 1 for a patient who underwent deep vein puncture (data for deep vein puncture treatment is 1), 0 for a patient who did not undergo deep vein puncture (data for deep vein puncture treatment is 0); 1 for a patient who received nasal cannula oxygen therapy (data for nasal cannula oxygen therapy is 1), 0 for a patient who did not receive nasal cannula oxygen therapy (data for nasal cannula oxygen therapy is 0); 1 for a patient who received carbapenem antibiotic therapy (data for carbapenem antibiotic therapy is 1), 0 for a patient who did not receive carbapenem antibiotic therapy (data for carbapenem antibiotic therapy is 0).

[0071] In one specific implementation, the module for calculating the risk of delirium in a patient pre-stores data based on existing databases, including age, fever data, chest pain data, diarrhea data, vomiting data, history of dementia, history of coronary heart disease, history of renal insufficiency, history of cancer, blood oxygen saturation, Glasgow Coma Scale (GCS) score, MEWS score, midazolam treatment data, deep vein puncture treatment data, nasal cannula oxygen therapy treatment data, and carbapenem antibiotic treatment data. This data is fitted using logistic regression to calculate the probability of delirium risk in a patient. The formula is pre-stored and can be used to calculate the probability of delirium in any patient.

[0072] During the calculation, this pre-stored formula uses data collected by the data acquisition module, including patient age, fever data, chest pain data, diarrhea data, vomiting data, history of dementia data, history of coronary heart disease data, history of renal insufficiency data, history of tumor data, blood oxygen saturation, GCS score, MEWS score, midazolam treatment data, deep vein puncture treatment data, nasal cannula oxygen therapy treatment data, and carbapenem antibiotic treatment data, to calculate the patient's risk probability of delirium.

[0073] Furthermore, this application constructs a specific formula for calculating the probability of delirium risk in patients, as shown in Formula 1 below: (Formula 1) Furthermore, in the above formula, The probability of a patient developing delirium, among which, These are the initial parameters. For the patient's age, Its corresponding regression coefficient; For fever data, Its corresponding regression coefficient; Data for chest pain; Its corresponding regression coefficient; Data for diarrhea; Its corresponding regression coefficient; Data on vomiting; Its corresponding regression coefficient; For dementia history data, Its corresponding regression coefficient; For data on past history of coronary heart disease, Its corresponding regression coefficient; Data on a history of renal insufficiency Its corresponding regression coefficient; For data on past cancer history, Its corresponding regression coefficient; Blood oxygen saturation Its corresponding regression coefficient; GCS score, Its corresponding regression coefficient; Rate MEWS Its corresponding regression coefficient; Data on midazolam treatment. Its corresponding regression coefficient; Data for deep vein puncture treatment. Its corresponding regression coefficient; Data on nasal cannula oxygen therapy. Its corresponding regression coefficient; Data on carbapenem antibiotic treatment. The corresponding regression coefficients are used to calculate the probability of a patient developing delirium based on the above variables and their corresponding regression coefficients.

[0074] In one specific implementation method It can be any value selected from 0.022 to 0.032, preferably 0.027; It can be any value selected from -0.686 to -0.458, preferably -0.572; It can be any value selected from -0.584 to -0.390, preferably -0.487; It can be any value selected from -1.256 to -0.838, preferably -1.047; It can be any value selected from -0.840 to -0.560, preferably -0.700; It can be any value selected from 0.210 to 0.316, preferably 0.263; It can be any value selected from -1.452 to -0.968, preferably -1.21; It can be any value selected from 1.036 to 1.554, preferably 1.295; It can be any value selected from 1.497 to 2.245, preferably 1.871; It can be any value selected from -0.043 to -0.029, preferably -0.036; It can be any value selected from -0.518 to -0.346, preferably -0.432; It can be any value selected from 0.123 to 0.185, preferably 0.154; It can be any value selected from 1.654 to 2.482, preferably 2.068; It can be any value selected from 0.654 to 0.980, preferably 0.817; It can be any value selected from 0.694 to 1.040, preferably 0.867; It can be any value selected from 0.710 to 1.064, preferably 0.887.

[0075] Furthermore, the system described in this application also includes a grouping module, in which default delirium risk grouping parameters are pre-stored, and the calculated probability of delirium risk in the subject is determined based on these grouping parameters. Subjects were grouped to assess their risk of delirium.

[0076] Example Example 1: Patient Screening and Experimental Data Collection Patients who visited the Emergency and Resuscitation Room of Peking University Third Hospital between May and November 2024 were included as the modeling group. During the study, patient information was strictly encrypted; patient names and medical record numbers were encoded to protect patient privacy.

[0077] Inclusion criteria: 1. Age ≥ 18 years; 2. Patients in the emergency room; 3. Observation time ≥ 24 hours.

[0078] Exclusion criteria: 1. Clinical data is severely lacking; 2. History of mental illness; 3. Acute cerebrovascular disease and traumatic brain injury, etc.; 4. Individuals who are unable to communicate and cooperate in completing the delirium assessment due to coma, severe dementia, or hearing, vision, or language impairments; 5. Pregnant patients.

[0079] Ultimately, 928 patients were included as the modeling dataset.

[0080] The data collection tool used in this application was designed jointly by a team of senior emergency medicine experts and clinical epidemiologists to comprehensively collect potential predictive factors for delirium. The study data collection form includes the following dimensions: (1) Basic Information: Patient ID, gender, age, chief complaint upon presentation including chest pain, abdominal pain, chest tightness, dyspnea, fever, syncope, fatigue, palpitations, hematemesis, melena / hemorrhagic stool, altered consciousness, headache, vomiting, trauma, cough, sputum production, hemoptysis, diarrhea; past medical history including dementia, cerebral infarction, epilepsy, cerebral hemorrhage, Parkinson's disease, schizophrenia, coronary heart disease, myocardial infarction, heart failure, hypertension, diabetes, cirrhosis, renal insufficiency, chronic obstructive pulmonary disease (COPD), history of alcoholism, history of long-term use of sedatives or antidepressants, history of tumors, etc., and time of admission to the emergency room. The above basic information is from the first emergency room visit medical record in the electronic medical record system.

[0081] (2) Physical examination: heart rate, respiratory rate, body temperature, systolic blood pressure, diastolic blood pressure, blood oxygen saturation, Glasgow Coma Scale (GCS), the assessment of which includes eye opening response (GCS-E), verbal response (GCS-V) and motor response (GCS-M) as well as the total GCS score, Richmond Agitation-Sedation Scale (RASS) score, and Modified Early Warning Score (MEWS). The above basic information is obtained from the first medical record of emergency visit in the electronic medical record system.

[0082] (3) Laboratory test results: including white blood cell count, hemoglobin, hematocrit, platelet count, creatinine, blood urea nitrogen, aspartate aminotransferase, alanine aminotransferase, albumin, total bilirubin, lactate dehydrogenase, sodium, potassium, calcium, magnesium, blood glucose, blood urea nitrogen, creatinine, whole blood troponin I, N-terminal B-type natriuretic peptide precursor, pH value, oxygen partial pressure, carbon dioxide partial pressure, lactate, procalcitonin, and D-dimer quantification. The above results are the first laboratory test results of the patient when they are admitted to the emergency room.

[0083] (4) Imaging data: The results of head imaging examinations (CT or MRI) include white matter lesions and age-related brain changes; the above head imaging data are derived from the first head CT / MRI examination report after the patient's emergency visit, and are completed by the radiology department of each research center.

[0084] (5) Treatment-related data: including sedatives and analgesics (midazolam, diazepam, dexmedetomidine, olanzapine, haloperidol), antibiotics (quinolones, carbapenems, penicillins, cephalosporins, tetracyclines, macrolides, aminoglycosides, etc.), glucocorticoids, respiratory support methods (nasal cannula, high-flow nasal cannula, oxygen reservoir mask, non-invasive ventilator, invasive ventilator), and whether central venous access was performed. Regarding the time window for collecting treatment data, for patients with delirium, treatment information before the first positive assessment of CAM-ICU should be collected, but not exceeding the first 24 hours after admission to the emergency resuscitation room; for patients without delirium, treatment data should be collected within the first 24 hours after admission to the emergency resuscitation room.

[0085] Example 2: Selecting prediction parameters from collected patient data for inclusion in the prediction system LASSO regression was used to screen a total of 78 variables, including 30 continuous variables (age, heart rate, diastolic blood pressure, systolic blood pressure, respiratory rate, oxygen saturation, body temperature, Glasgow Coma Scale (GCS) score, Glasgow Coma Scale (GCS) eye-opening response, Glasgow Coma Scale (GCS) verbal response, Glasgow Coma Scale (GCS) motor response, white blood cell count, hematocrit, hemoglobin, platelet count, calcium ion level, potassium ion level, sodium ion level, alanine aminotransferase (ALT), aspartate aminotransferase (AST), creatinine, blood urea nitrogen (BUN), albumin, total bilirubin, lactate dehydrogenase (LDH), blood glucose, lactate, pH, arterial oxygen partial pressure, and revised early warning score) and 48... The following categorical variables were selected for screening: altered consciousness, hematemesis, palpitations, syncope, fever, dyspnea, abdominal pain, chest pain, gender, diarrhea, hemoptysis, cough, vomiting, headache, injury, history of chronic obstructive pulmonary disease, history of alcoholism, history of renal insufficiency, history of cirrhosis, history of heart failure, history of myocardial infarction, history of coronary heart disease, history of Parkinson's disease, history of cerebral hemorrhage, history of epilepsy, history of cerebral infarction, history of dementia, history of diabetes, history of hypertension, history of tumors, RASS score, use of midazolam, use of glucocorticoids, use of quinolone antibiotics, use of penicillin, use of macrolide antibiotics, deep vein puncture, nasal cannula oxygen therapy, high-flow nasal cannula oxygen therapy, oxygen mask, non-invasive ventilation, mechanical ventilation, vasopressors, use of non-benzodiazepine sedatives, use of dexmedetomidine, use of carbapenems, use of cephalosporins, and use of aminoglycosides.

[0086] Figure 1 The graph shows the coefficient changes of 78 candidate variables under different regularization parameters. The horizontal axis represents the logarithm of the regularization parameter λ, and the vertical axis represents the standardized regression coefficients. Each colored curve represents the coefficient change trajectory of a variable, and the numbers above the graph indicate the number of non-zero variables at the corresponding λ value. As the λ value increases, the model complexity decreases, and the coefficients of some variables gradually shrink to zero, thus being excluded from the model. Ultimately, only the core variables most critical for predicting delirium are retained.

[0087] Figure 2 shows the relationship between the 10-fold cross-validation error (CVM) and the Log(λ) value. The vertical line segments represent the standard error range corresponding to each λ value. When the minimum CVM corresponding to the λ value is selected, Log(λ_min) = -4.629 (red dashed line in the figure), the model retains 40 variables; while when a more simplified model within one standard error range is selected (Log(λ_1se) = -3.606 (blue dashed line in the figure), the model retains 16 variables.

[0088] As shown in the figure, although the CVM at Log(λ_1se) is slightly higher than the minimum point, it is still within the standard error range of the minimum point, indicating that there is no statistically significant difference between the two models. Considering the feasibility of variable collection in an emergency clinical setting and the model selection principle based on the 1SE rule, this application selects 16 variables at λ_1se as the optimal scheme for delirium prediction, which ensures both prediction accuracy and meets the simplicity requirements of clinical application.

[0089] The 16 variables selected by LASSO regression at λ_1se are: age, fever, chest pain, diarrhea, vomiting, history of renal insufficiency, history of cancer, history of coronary heart disease, blood oxygen saturation, GCS total score, GCS exercise response score, MEWS score, central venous access, nasal cannula oxygen therapy, midazolam use, and carbapenem antibiotic use. Considering the high collinearity between the GCS total score and the GCS exercise response score, we retained the more comprehensive GCS total score and excluded the GCS exercise response score. Additionally, dementia had a low incidence in this cohort (1.1%) and was not selected by the LASSO algorithm, but it is widely recognized as an independent risk factor for delirium, ranking as the second most important risk factor by emergency physicians in a national survey (score 3.78±1.00). To enhance the clinical effectiveness and generalization ability of the model, we decided to forcibly include it in the model. After the above calibration, a predictive model containing 16 feature variables was finally formed. It should be noted that the variable selection strategy in this application is mainly based on LASSO regression. The subsequent multivariate logistic regression is only used to show the effect size and statistical characteristics of the selected variables. The final inclusion of variables is based on the LASSO regression results rather than the traditional p-value standard.

[0090] Example 3: Determining the Prediction System Model The data from 928 patients in the modeling group were randomly divided into a training set (650 cases) and a test set (278 cases) at a ratio of 70% and 30%, respectively. Using the 16 feature variables selected by the LASSO regression as modeling parameters, four machine learning methods were used to build the model.

[0091] Mainly includes: (1) Logistic Regression: The linear combination is mapped to probability values ​​using the Sigmoid function. P(Y=1|X) = 1 / (1+e^(-β0-β1X1-...-β p X p ) L2 regularization (regularization strength C=1.0) was used to prevent overfitting, and the optimization algorithm was a limited-memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS) solver, with a maximum number of iterations set to 100. The regression coefficients directly reflect the intensity and direction of the influence of each risk factor on the occurrence of delirium, facilitating clinical interpretation.

[0092] (2) Extreme gradient boosting: By iteratively constructing multiple decision trees to correct prediction errors, it can effectively capture the nonlinear relationships and interactions between risk factors. Default parameter configuration is used to maintain the algorithm's versatility.

[0093] (3) Classification Enhancement: Using ordered target encoding technology, it can natively process categorical variables without preprocessing steps. With 100 iterations, a maximum tree depth of 10, and a learning rate of 0.1, it is particularly suitable for processing clinical data containing a large number of categorical variables.

[0094] (4) Random Forest: The number of decision trees is set to 100, the maximum depth is limited to 3, and Gini impurity is used as the splitting criterion. This algorithm can integrate multidimensional clinical data and assess the relative importance of different factors.

[0095] Among the four models mentioned above, this system ultimately selected the logistic regression model as the core algorithm for calculating the risk of delirium in patients. Although ensemble algorithms such as extreme gradient boosting, classification boosting, and random forests had slightly higher area under the receiver operating characteristic (AUC) than the logistic regression model in the initial training and internal validation using single-center modeling data, due to their superior fitting of complex nonlinear relationships, the logistic regression model showed a significant advantage in discrimination during subsequent independent external validation (including the Peking University First Hospital cohort and the international MIMIC-IV database). Its external validation set AUC reached 0.825 and 0.784, respectively, both consistently outperforming the other three algorithms at the same time. This phenomenon indicates that complex algorithms are prone to overfitting when dealing with highly heterogeneous clinical data from the emergency department. The logistic regression model with L2 regularization used in this application demonstrates superior generalization ability and data robustness, effectively overcoming the performance degradation risk caused by differences in cross-center and cross-system data distribution. Furthermore, the calibration curves from external validation also confirm that the logistic regression model has the highest consistency between its predicted probabilities and actual observed results, exhibiting the best calibration performance. More importantly, decision support systems in emergency and critical care settings demand high clinical transparency from their models. Based on the aforementioned generalization advantages and interpretability considerations, we further extracted and analyzed the internal parameters of this logistic regression model.

[0096] Considering clinical interpretability, we further analyzed the internal parameters of the logistic regression model.

[0097] As shown in Table 1 below, the logistic regression model results show that midazolam use, history of cancer, and history of renal insufficiency are the main risk factors, while history of coronary heart disease, diarrhea, and fever are protective factors; increasing age is associated with an increased risk of delirium.

[0098] Table 1. Multivariate Logistic Regression Analysis

[0099] Note:* P <0.05.

[0100] Figure 3This paper presents a nomogram predicting delirium in critically ill emergency patients, built based on a logistic regression model. This tool integrates 16 predictor variables into a visualized clinical decision support system, facilitating rapid bedside risk assessment. The nomogram includes an integral scale and scoring axes for each predictor variable, with each variable assigned a weight based on its regression coefficient. Continuous variables (age, oxygen saturation, MEWS score) use a continuous scale, while categorical variables are presented in a binary format. In use, clinicians determine values ​​on each variable axis based on the patient's condition, project them upwards onto the integral scale to obtain scores, sum all scores to obtain a total score, and finally read the probability of delirium occurrence from the bottom axis using the total score.

[0101] Comparative Example 1 To further validate the clinical application value of the model in this application, we selected a logistic regression model that combines good performance and clinical interpretability, and compared its predictive efficacy with the existing E-PRE-DELIRIC scoring model. Figure 4 The ROC curves of the logistic regression model in this application are compared with those of the E-PRE-DELIRIC model on a randomized single-split validation set. The results show that the AUC of the delirium prediction model (M2) constructed in this application is 0.851 (95% CI: 0.814–0.887), while the AUC of the E-PRE-DELIRIC scoring model (M1) is 0.692 (95% CI: 0.647–0.737). The DeLong test shows that the difference in AUC between the two models is statistically significant (Z = -6.788, p < 0.001). The model in this application is not inferior to the E-PRE-DELIRIC scoring model.

[0102] Comparative Example 2 This section aims to conduct external validation using a cohort of critically ill patients in the emergency department of Peking University First Hospital to evaluate the model's generalization ability.

[0103] This application includes patients who visited the Emergency and Resuscitation Room of Peking University First Hospital between May and November 2024 as an independent external validation cohort in China. The initial sample size for the model was n=928, with this portion calculated as n=93 cases (10% of the total). According to general guidelines, the external validation set sample size should not be less than 100 cases. Inclusion and exclusion criteria were the same as those used in the modeling group described above, and the Emergency and Resuscitation Room patient data were collected using the same structured data collection method as the modeling group described above. The main objective is to validate the predictive performance of the logistic regression method described in this application and three other machine learning algorithms on an independent dataset. Based on the model evaluation index system described above, performance results were evaluated using multi-dimensional indicators, including accuracy, sensitivity, specificity, positive predictive value, Youden index, and AUC.

[0104] On an independent external validation set of the emergency and critically ill patient cohort at Peking University First Hospital, the receiver operating characteristic (ROC) curve analysis of the logistic regression model described in this application and three other prediction models is as follows: Figure 5 As shown. From Figure 5 As can be seen, the AUCs of the four methods on the validation set are 0.825 (95% CI: 0.745-0.898), 0.799 (95% CI: 0.713-0.878), 0.799 (95% CI: 0.713-0.873), and 0.794 (95% CI: 0.705-0.874), respectively.

[0105] The optimal cutoff value, sensitivity, specificity, and accuracy of each model are detailed in Table 2.

[0106] Table 2 Performance metrics of the four prediction models on the validation set at Peking University First Hospital

[0107] As shown in Table 2, the overall predictive performance of the logistic regression model on the independent external validation set is superior to the other three ensemble algorithms. In independent external validation, the logistic regression model outperforms the other three algorithms. From the perspective of model discriminative ability, the logistic regression AUC reaches 0.825, higher than extreme gradient boosting (0.799), classification boosting (0.799), and random forest (0.794), demonstrating better cross-center applicability. Furthermore, logistic regression not only has the highest accuracy (0.770), but also the most balanced ratio of missed diagnoses to false diagnoses. Moreover, the optimal diagnostic threshold of logistic regression (0.503) is closer to the intuitive value of 0.5, and its output risk probability is more consistent with the clinical experience of emergency physicians. Therefore, this system ultimately uses it as the core prediction algorithm.

[0108] The logistic regression model developed for this application was directly compared with the internationally recognized E-PRE-DELIRIC scoring system on the validation set of Peking University First Hospital. The results are as follows: Figure 6As shown, the AUC of the delirium logistic regression prediction model constructed in this application is 0.835 (95% CI: 0.762-0.908), while the AUC of the E-PRE-DELIRIC scoring system is 0.717. Statistical analysis based on the DeLong test indicates a statistically significant trend in the difference in AUC values ​​between the two models (p = 0.052). The Net Reclassification Improvement Index (NRI) of this application's model is 0.603 (95% CI: 0.288-0.917, p < 0.001), and the Integrated Discriminant Improvement Index (IDI) is 0.137 (95% CI: 0.041-0.232, p = 0.005). Both of these indicators show statistically significant differences, indicating that the model in this application has a significantly improved ability to correctly classify patient risk levels compared to E-PRE-DELIRIC.

[0109] Comparative Example 3 This application uses the MIMIC-IV database as a foreign data validation set. MIMIC-IV is a large medical database released by the MIT Computational Physiology Laboratory. This database organizes data using a modular approach, primarily divided into two modules: hospital data (hospital data) and ICU (intensive care unit data). The database records detailed patient demographic characteristics, admission information, intra-hospital transfer records, laboratory test results, microbial cultures, medical orders, medication management, hospital billing information, and more. This application tracks patients admitted from the emergency department and transferred to the ICU within a short period to assess the incidence of delirium in critically ill emergency patients. All researchers in this application have completed Collaborative Institutional Training Initiative (CITI) training and obtained the corresponding certification. PhysioNet was used as the validation dataset. Access to MIMIC-IV and MIMIC-ED was obtained, and the PhysioNet Credentialed Health Data License 1.5.0 was strictly followed during data acquisition and analysis. MIMIC-IV received approval from the Institutional Review Boards of Beth Israel Deaconess Medical Center (BIDMC) and the Massachusetts Institute of Technology (MIT), and data use complied with ethical requirements. Considering that the MIMICIV database is primarily based on intensive care unit data, and that there are certain differences in treatment models between China and the United States, appropriate adjustments were made to the study subjects in this application to better match the definition of critically ill emergency patients in this application.

[0110] The specific inclusion criteria are as follows: 1. Emergency room admission; 2. Select patients who are transferred to the ICU within 24 hours of admission to the emergency department; 3. Data collection period: from 2008 to 2019.

[0111] Exclusion criteria include: 1. Clinical data is severely lacking; 2. Patients with acute cerebrovascular disease and traumatic brain injury; 3. Patients with missing key predictor or outcome variables.

[0112] MIMIC-IV data extraction employed SQL structured query language to extract parameters. The database version used was MIMIC-IV v3.0. A multi-module collaborative strategy was used for MIMIC-IV data extraction, obtaining demographic characteristics and past medical history from the hospital module (through ICD coding matching), and monitoring and treatment information and delirium assessment results from the ICU module. A longitudinal data chain was constructed using subject_id, hadm_id, and stay_id as association identifiers, and patients transferred from ED to ICU, as well as those with an interval of <24 hours, were screened. For patients with delirium, data were collected before the first positive assessment in CAMICU, but not exceeding the first 24 hours after ICU admission. For patients without delirium, data were collected within the first 24 hours after ICU admission. Clinical data extraction followed the principle of "first measurement priority upon admission": vital signs were prioritized using the vitalsign table, laboratory results were used using the lab table, and the GCS score employed a hybrid strategy from both ED and ICU modules. Delirium assessment used the CAM-ICU records in the chartevents table of the ICU module, defining a positive score as the occurrence of delirium. For vital signs data, the initial admission record was used first; if data was missing, the first measurement after ICU admission was used as a substitute. Laboratory test results were collected from the first results after admission. The study endpoint was the occurrence of delirium, defined as a positive CAM-ICU score. All data extraction processes underwent cross-validation to ensure accuracy.

[0113] The MIMIC-IV validation study compared 16 predictor characteristics between the delirium and non-delirium groups, revealing significant differences between the two groups. Patients in the delirium group were older, and the incidence of tumors, renal insufficiency, dementia, and fever was significantly higher than in the non-delirium group. The non-delirium group had a higher incidence of chest pain, diarrhea, and vomiting. The delirium group had lower Glasgow Coma Scale (GCS) scores, higher MEWS scores, and slightly lower oxygen saturation. Regarding treatment, the delirium group had significantly higher rates of midazolam use, carbapenem antibiotic use, nasal cannula oxygen therapy, and central venous catheterization than the non-delirium group. Detailed data are shown in the table below: Table 3. Comparison of characteristics of 16 predictors in the delirium prediction model in the MIMIC-IV validation set.

[0114] The four models described above were validated on the external validation dataset of the MIMIC-IV database. The results show that the different models perform well on the external data. Detailed results are as follows: Figure 7 As shown, the AUCs of the four machine learning models are: Logistic Regression 0.784 (95% CI: 0.771-0.798), Extreme Gradient Boosting 0.731 (95% CI: 0.716-0.746), Classification Boosting 0.738 (95% CI: 0.724-0.751), and Random Forest 0.758 (95% CI: 0.744-0.771).

[0115] The sensitivity and specificity of each model are shown in the table below: Table 4 External validation performance metrics of the MIMIC-IV model

[0116] Note: AUC: Area under the ROC curve; the optimal threshold is determined based on the principle of maximizing the Youden Index (sensitivity + specificity - 1).

[0117] As shown in Table 4, the logistic regression model outperformed the other three ensemble algorithms in overall predictive performance on the MIMIC-IV external validation dataset. In terms of discriminant performance metrics, the logistic regression model exhibited the highest AUC (0.784) and accuracy (0.685), demonstrating excellent predictive performance. Regarding clinical applicability, while some ensemble models showed slightly higher sensitivity, their specificity was generally low (all below 0.6), easily leading to excessive false positives in practical applications. In contrast, the logistic regression model achieved a better balance between sensitivity (0.801) and specificity (0.645), effectively reducing the risk of misdiagnosis while ensuring a high detection rate. In conclusion, the logistic regression model demonstrated the most robust comprehensive application value in this cross-center external validation, and therefore was selected as the core prediction algorithm for this system. Figure 8 The calibration curves of the model described in this application and three other machine learning models on the MIMIC-IV validation set are shown to demonstrate the consistency between the predicted probabilities and actual observations. Logistic regression's calibration curve is closest to the ideal calibration line, exhibiting the best calibration performance. Extreme gradient boosting and classification boosting both fall below the ideal calibration line, overestimating the probability of events. Random forest overestimates risk in the low-probability range but shows relatively good calibration performance in the medium-to-high-probability range.

[0118] Example 4: Delirium Risk Prediction System for Critically Ill Patients in the Emergency Department as Described in this Application Based on the best-performing logistic regression model in internal and external validation of this application, the research team successfully developed and deployed the "ED-DPMC Risk Prediction System for Emergency Critically Ill Patients." This system is a fully functional web application. It provides a structured data entry interface integrating all 16 predictors. An example of the system's user interface is shown below. Figure 9 As shown. After the user submits data, the system's backend prediction engine calls the fixed logistic regression model in real time to instantly calculate the individualized risk probability of the patient developing delirium (presented as a value in the range of 0%–100%). The prediction results are presented together as risk probability and corresponding risk level (e.g., divided into low risk, medium risk, and high risk according to preset thresholds). The final web version of this program is published at https: / / shiny-cucurucho-9537b9.netlify.app, and its built-in prediction algorithm is trained based on the modeling cohort data of 928 patients in this application.

[0119] Although the embodiments of this application have been described above in conjunction with the accompanying drawings, this application is not limited to the specific embodiments and application fields described above. The specific embodiments described above are merely illustrative and instructive, not restrictive. Those skilled in the art can make many other forms based on the guidance of this specification and without departing from the scope of protection of the claims of this application, and these are all within the scope of protection of this application.

Claims

1. A system for predicting the risk of delirium in patients, comprising: The data acquisition module is used to acquire patient data including age, fever, chest pain, diarrhea, vomiting, history of dementia, history of coronary heart disease, history of renal insufficiency, history of cancer, blood oxygen saturation, Glasgow Coma Scale (GCS) score, Meniere's Disease Scale (MEWS) score, midazolam treatment data, deep vein puncture treatment data, nasal cannula oxygen therapy treatment data, and carbapenem antibiotic treatment data; as well as... The module for calculating the risk of delirium in patients is used to perform calculations on the data acquired by the data acquisition module to determine the probability of delirium in patients. .

2. The system according to claim 1, wherein, Age, blood oxygen saturation, GCS score, and MEWS score are continuous variables.

3. The system according to claim 1, wherein, The data on fever, chest pain, diarrhea, vomiting, history of dementia, history of coronary heart disease, history of renal insufficiency, history of tumors, treatment with midazolam, treatment with deep vein puncture, treatment with nasal cannula oxygen therapy, and treatment with carbapenems are binary variables.

4. The system according to claim 3, wherein, If the patient's body temperature is greater than or equal to 37.3°C, the fever data is 1; if the patient's body temperature is less than 37.3°C, the fever data is 0. The patient has chest pain symptoms, and the chest pain data is 1; the patient has no chest pain symptoms, and the chest pain data is 0. If a patient has diarrhea symptoms, the diarrhea data is 1; if a patient does not have diarrhea symptoms, the diarrhea data is 0. If the patient has vomiting symptoms, the vomiting data is 1; if the patient does not have vomiting symptoms, the vomiting data is 0. The patient has a history of dementia, and the data for the history of dementia is 1; the patient has no history of dementia, and the data for the history of dementia is 0. The patient has a history of coronary heart disease, and the data for the history of coronary heart disease is 1; the patient has no history of coronary heart disease, and the data for the history of coronary heart disease is 0. The patient has a history of renal insufficiency, and the data for the history of renal insufficiency is 1; the patient has no history of renal insufficiency, and the data for the history of renal insufficiency is 0. The patient has a history of cancer, and the cancer history data is 1; the patient has no history of cancer, and the cancer history data is 0. The patient is being treated with midazolam, and the midazolam treatment data is 1; the patient is not being treated with midazolam, and the midazolam treatment data is 0. The patient has undergone deep vein puncture treatment, and the deep vein puncture treatment data is 1; the patient has not undergone deep vein puncture treatment, and the deep vein puncture treatment data is 0. The patient received nasal cannula oxygen therapy, and the nasal cannula oxygen therapy data is 1; the patient did not receive nasal cannula oxygen therapy, and the nasal cannula oxygen therapy data is 0. The patient received carbapenem antibiotic treatment, and the carbapenem antibiotic treatment data is 1; the patient did not receive carbapenem antibiotic treatment, and the carbapenem antibiotic treatment data is 0.

5. The system according to claims 1-4, wherein, The module for calculating the risk of delirium in patients pre-stores data based on existing databases, including age, fever, chest pain, diarrhea, vomiting, history of dementia, history of coronary heart disease, history of renal insufficiency, history of cancer, blood oxygen saturation, Glasgow Coma Scale (GCS) score, MEWS score, midazolam treatment data, deep vein puncture treatment data, nasal cannula oxygen therapy data, and carbapenem antibiotic treatment data. This data is fitted using logistic regression to calculate the probability of delirium risk in patients. The formula.

6. The system according to claim 5, wherein, The formula is: in, These are the initial parameters. For the patient's age, Its corresponding regression coefficient; For fever data, Its corresponding regression coefficient; Data for chest pain; Its corresponding regression coefficient; Data for diarrhea; Its corresponding regression coefficient; Data on vomiting; Its corresponding regression coefficient; For dementia history data, Its corresponding regression coefficient; For data on past history of coronary heart disease, Its corresponding regression coefficient; Data on a history of renal insufficiency Its corresponding regression coefficient; For data on past cancer history, Its corresponding regression coefficient; Blood oxygen saturation Its corresponding regression coefficient; GCS score, Its corresponding regression coefficient; Rate MEWS Its corresponding regression coefficient; Data on midazolam treatment. Its corresponding regression coefficient; Data for deep vein puncture treatment. Its corresponding regression coefficient; Data on nasal cannula oxygen therapy. Its corresponding regression coefficient; Data on carbapenem antibiotic treatment. Its corresponding regression coefficient.

7. The system according to claim 6, wherein, It can be any value selected from 0.022 to 0.032; It can be any value selected from -0.686 to -0.458; For any value selected from -0.584 to -0.390; It can be any value selected from -1.256 to -0.838; It can be any value selected from -0.840 to -0.560; It can be any value selected from 0.210 to 0.316; It can be any value selected from -1.452 to -0.968; It can be any value selected from 1.036 to 1.554; It can be any value selected from 1.497 to 2.245; It can be any value selected from -0.043 to -0.029; It can be any value selected from -0.518 to -0.346; It can be any value selected from 0.123 to 0.185; It can be any value selected from 1.654 to 2.482; It can be any value selected from 0.654 to 0.980; It can be any value selected from 0.694 to 1.040; It can be any value selected from 0.710 to 1.

064.

8. The system according to claim 7, wherein, The preferred value is 0.027; The preferred value is -0.572; The preferred value is -0.487; The preferred value is -1.047; The preferred value is -0.700; The preferred value is 0.263; The preferred value is -1.21; The preferred value is 1.295; The preferred value is 1.871; The preferred value is -0.036; The preferred value is -0.432; The preferred value is 0.154; The preferred value is 2.068; The preferred value is 0.817; The preferred value is 0.867; The preferred value is 0.

887.

9. The system according to claim 1, further comprising: The grouping module contains pre-stored default delirium risk grouping parameters, and based on these parameters, it assigns the calculated probability of delirium risk to the subject. Subjects were grouped to assess their risk of delirium.

Images