Method for constructing a risk prediction model for post-stroke delirium

Abstract

The present invention discloses a method for constructing a risk prediction model for post-stroke delirium, which relates to the technical field of big data processing. By integrating different types of research and individual data of patients, a comprehensive risk factor data set D is established, covering key information of general information of patients, past medical history and symptoms after admission. This multi-dimensional data collection method avoids neglecting individual characteristics of patients and solves the defect that individual differences cannot be accurately adapted in existing models. By calculating the risk factor association matrix A and the matching factor Mm, it is ensured that the association between different symptoms and risk factors is accurately captured and analyzed, especially by screening out the gradient marking state, the prediction flexibility and accuracy of the model are further enhanced. The complication manifestation set M_Fi is matched with the patient data, and the personalized delirium risk Pd is calculated. By comparing with the preset risk assessment threshold Td, a personalized risk assessment strategy is generated.

Description

Technical Field

[0001] The present invention relates to the technical field of big data processing, and in particular to a method for constructing a risk prediction model for post-stroke delirium. Background Art

[0002] With the development of modern medical technology, personalized medicine has become a new trend. Personalized risk assessment is particularly important in the rehabilitation process of neurological diseases such as stroke. According to the rehabilitation effects of different patients, the risk prediction model can identify high-risk patients in advance and conduct personalized management.

[0003] The invention of Chinese patent application number CN202310055764.X discloses a method for constructing a cancer recurrence risk prediction model. By integrating single-cell and whole transcriptome sequencing data sets, prognosis-related genes that affect immunotherapy response are identified, and a prognostic risk model is constructed. This is beneficial to the formulation of application manuals for different tumor diagnostic biomarkers and the design of targeted drugs, providing new ideas for the clinical diagnosis and treatment of tumors.

[0004] It can be seen that although there are many risk prediction models available and many post-stroke delirium risk assessment models are used in clinical practice, most models still use fixed risk factor weights and lack adjustments for individual differences. Although this one-size-fits-all approach can give a rough risk assessment, it is difficult to take into account the differences in individual characteristics such as age and medical history among different patients. These models usually rely on common data sets and fixed association rules, ignoring the impact of the patient's unique medical history and physiological state on the prediction results, resulting in insufficient accuracy of model predictions. For example, older patients and younger patients have obvious differences in recovery after stroke, but many existing models do not fully reflect these differences. Summary of the invention

[0005] In view of the deficiencies of the prior art, the present invention provides a method for constructing a risk prediction model for post-stroke delirium, which solves the problems mentioned in the background technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for constructing a risk prediction model for post-stroke delirium, comprising the following steps:

[0007] S1, by collecting cases of risk factors for post-stroke delirium in stroke patients, including control studies, cohort studies, cross-sectional studies, general information of patients, medical history, symptoms after admission, blood indicators and scores of medical staff, integrating documents and performing big data retrieval to find relevant words related to risk factors for post-stroke delirium in relevant patients, and then extracting word occurrence frequency and related sequence information to form the patient's risk factor data set D;

[0008] S2, by performing mutual correlation analysis on the risk factor data set D, forming risk factor data Fi marked with correlation relationships, and reorganizing the risk factor data Fi to obtain a risk factor correlation matrix A;

[0009] S3, traverse and calculate each risk factor data Fi of the risk factor association matrix A, obtain the matching factor Mm between the risk factor data Fi in the association relationship, match it with the preset matching degree screening threshold PZ, and output the gradient marking state result of the matching factor Mm of each risk factor data Fi;

[0010] S4, screening according to the gradient marking state result of the matching factor Mm of each risk factor data Fi, and outputting the complication manifestation set M_Fi of each risk factor data Fi;

[0011] S5, summarize the complication manifestation set M_Fi, and calculate the patient's delirium risk Pd by using a nonlinear calculation formula;

[0012] S6, obtaining the delirium risk assessment strategy for the current patient by matching the obtained delirium risk Pd with the preset risk assessment threshold Td.

[0013] Preferably, said S1 includes S11, S12 and S13;

[0014] S11, select stroke patients who meet the inclusion criteria in the hospital's neurology and neurosurgery departments as research subjects, and collect research information on risk factors for post-stroke delirium in stroke patients, including control studies, cohort studies and cross-sectional studies. Combined with case information, general information, past medical history, symptoms after admission, blood indicators and medical staff scores are collected to form the initial patient data set Draw={P1, P2, ..., Pn} for preliminary retrieval, where P1 to Pn represent case patient 1 to case patient n.

[0015] Preferably, S12, extracting key words w related to delirium risk from the initial patient data set Draw, and counting the occurrence frequency of the key words w, calculating the frequency fi of the key words w in the case patient data through big data processing technology, and marking the sequence information S of the occurrence of the related key words, the position and order of the key words in the patient's case information, and forming a key word frequency set Q={f1, f2, ..., fi};

[0016] The frequency fi is obtained by the following calculation formula: ; In the formula, fi represents frequency, specifically represents the frequency of the keyword appearing in the initial patient data set Draw, n represents the length of the initial patient data set Draw, specifically represents the total number of case patients included, wm represents the mth keyword wm related to the risk of post-stroke delirium, specifically m represents the subscript information of the keyword w, the mth keyword wm is determined by the combination of wm, and then the mth keyword wm is located by wm, M represents the total number of keywords, specifically obtained by counting the total number of preset search keywords, Pj represents the jth case patient in the initial patient data set Draw, wherein the search keywords specifically include gender, age, previous stroke, previous cognitive impairment, hypertension, hyperlipidemia, diabetes, stroke type, smoking, drinking, atrial fibrillation, ischemic heart disease, heart failure, infection, intubation, medium cell ratio, lymphocyte ratio, fever, urinary incontinence, visual field loss, ADL life ability score, NIHSS score, aphasia and dysphagia keywords.

[0017] Preferably, S13, extracting sequence information S related to risk factors from the keyword frequency set Q, and simultaneously combining the first to mth keywords and the sequence information S where they appear into sequence features, and integrating them into a risk factor data set D;

[0018] The risk factor dataset D is obtained by the following integration method: ; Wherein, Sk represents the kth sequence information of the mth keyword wm.

[0019] Preferably, S2 includes S21 and S22;

[0020] S21, performing association analysis based on sequence information S on the mth keyword wm and the pth keyword wp related to the risk of post-stroke delirium according to the acquired risk factor dataset D, obtaining the association between the appearance patterns of the mth keyword wm and the pth keyword wp at different sequence information S positions, generating an association mark Rmp, and forming risk factor data Fi={(wm, wp, Rmp)} marked with the association relationship;

[0021] The association mark Rmp is obtained by the following calculation formula: ; Wherein, Rmp represents an association mark, specifically represents the association strength between the mth keyword and the pth keyword, Nm represents the total number of occurrences of the mth keyword, k represents the kth time that the mth keyword wm appears in the total number Nm, Np represents the total number of occurrences of the pth keyword, l represents the lth time that the pth keyword wp appears in the total number Np, represents the average sequence position of the key word wm, represents the average sequence position of the keyword wp, Swmk represents the sequence information of the keyword wm that appears for the kth time, and Swpl represents the sequence information of the keyword wp that appears for the lth time.

[0022] Preferably, S22, reorganizing the association information of the mth keyword wm and the pth keyword wp according to the generated risk factor data Fi marked with an association relationship, and constructing an association matrix A of associated words, specifically indicating the mutual association degree between the mth keyword wm and the pth keyword wp;

[0023] The association matrix A is specifically composed in the following manner: ; In the formula, each matrix element Rmp in the association matrix A represents the association strength between the keyword wm and the keyword wp, and the diagonal element Rmm=1 represents self-association, wherein the mth keyword wm and the pth keyword wp may be adjacent, or may not be adjacent, and may even be separated by multiple positions.

[0024] Preferably, the matching factor Mm is obtained by the following calculation formula: ; Wherein, Mm represents the matching factor, specifically the matching factor of the mth keyword wm, amp represents the element in the association matrix A, specifically the association strength between the mth keyword and the pth keyword, and Pp represents the frequency value of the pth keyword in the keyword frequency set Q;

[0025] The matching factor Mm is calculated by traversing the mth row of the association matrix A, that is, calculating the association strength between the key word wm and all other key words wp, and weighting the value of each key word according to these association strengths;

[0026] The gradient marking state result of the matching factor Mm of each risk factor data Fi is obtained by the following matching method: ; When the matching factor Mm≥matching degree screening threshold PZ, the gradient marking state result is obtained as the first gradient marking, and Label=1, indicating that the matching relationship between the mth keyword wm and the pth keyword wp in the risk factor data Fi is marked as 1, specifically indicating that the mth keyword wm and the pth keyword wp are of the first gradient association degree;

[0027] When the matching factor Mm is less than the matching degree screening threshold PZ, the gradient marking state result is obtained as the second gradient marking, and Label=2, indicating that the matching relationship between the mth keyword wm and the pth keyword wp in the risk factor data Fi is marked as 2, specifically indicating that the mth keyword wm and the pth keyword wp are of the second gradient association.

[0028] Preferably, the complication manifestation set M_Fi is obtained by integrating the risk factor data Fi based on the gradient marking status result, specifically by screening the pth keyword wp whose mth keyword wm and pth keyword wp have a correlation of the first gradient, thereby forming the complication manifestation set M_Fi of the risk factor data Fi with only the correlation of the first gradient.

[0029] Preferably, the patient's delirium risk Pd is calculated by using a nonlinear calculation formula, wherein the patient obtains general information, past medical history, symptoms after admission, blood indicators and medical staff scoring information through steps S1 to S4 to form the patient's physical sign data set P, and the physical sign data set P is matched with the complication manifestation set M_Fi to obtain the total number of complications Nmatch, and is calculated with the length value MFC of the complication manifestation set M_Fi to obtain the delirium risk Pd;

[0030] The delirium risk Pd is determined by Obtain the calculation formula.

[0031] Preferably, the current patient's delirium risk assessment strategy is obtained by the following matching method:

[0032] When the delirium risk Pd is less than the risk assessment threshold Td, the delirium risk assessment of the current patient is obtained as no risk strategy;

[0033] When the delirium risk Pd ≥ the risk assessment threshold Td, the delirium risk assessment of the current patient is obtained as a delirium risk strategy, the current patient case information is marked and retained, and a notification is sent to the relevant doctor for consultation;

[0034] When the delirium risk Pd ≥ twice the risk assessment threshold Td, the delirium risk strategy assessment of the current patient is obtained as the delirium warning risk strategy, the current patient case information is marked and retained, and a delirium warning risk strategy notification is sent to relevant doctors for processing.

[0035] The present invention provides a method for constructing a risk prediction model for post-stroke delirium, which has the following beneficial effects:

[0036] Through steps S1 to S6, the individualized data of patients are fully utilized, the accuracy of risk prediction is improved, the shortcomings caused by the one-size-fits-all problem in traditional models are solved, and the whole process from data collection, association analysis to risk assessment is realized. First, by integrating different types of research and individual data of patients, a comprehensive risk factor data set D is established, covering key information such as general information of patients, past medical history, and symptoms after admission. This multi-dimensional data collection method avoids ignoring individual characteristics of patients and solves the defect that the existing model cannot accurately adapt to individual differences. Then, by calculating the risk factor association matrix A and the matching factor Mm, the association between different symptoms and risk factors is ensured to be accurately captured and analyzed, especially by screening out the gradient marker state, which further enhances the prediction flexibility and accuracy of the model. The complication manifestation set M_Fi is matched with the patient data, and the personalized delirium risk Pd is calculated. By comparing it with the preset risk assessment threshold Td, a personalized risk assessment strategy is generated. This process ensures that the risk assessment of different patients is more targeted and avoids misjudgment and delay caused by ignoring individual differences in traditional models.

[0037] By integrating the patient data that meets the inclusion criteria, the initial patient data set Draw is generated, covering a variety of research data such as control studies, cohort studies, and cross-sectional studies, which provides a solid foundation for the subsequent delirium risk keyword extraction. The initial patient data set Draw is analyzed in detail to extract the key words w related to delirium risk, and the frequency of occurrence of each keyword fi and the sequence information S of the keyword in the case information are calculated. This enables this method to not only capture the overall frequency of the keyword, but also identify its specific distribution in different patient cases to form a keyword frequency set Q. Finally, by integrating the keywords and their sequence information S into the risk factor data set D, the temporal features and association patterns in the patient data can be effectively retained and refined, so that the risk assessment model can accurately analyze the personalized risk factors of each patient. The special advantage of this process is that it can further refine the risk assessment by introducing the temporal and association features in the patient cases, and improve the flexibility and dynamic prediction ability of the model.

[0038] Using the risk factor dataset D, the key words related to post-stroke delirium were subjected to association analysis based on sequence information. Not only the association relationship between the key words wm and wp was extracted, but also the occurrence patterns and mutual relationships of these key words at different positions were captured by calculating the association mark Rmp. This association analysis based on sequence information can identify the temporal association characteristics of risk factors, and the generated association mark Rmp can provide more accurate association information for subsequent model optimization. The risk factor data was reorganized by the association mark Rmp, and the association matrix A between the key words was constructed. The association strength between each keyword was refined, and the matching factor Mm of each risk factor data was generated by traversing the association matrix. By calculating the gradient marking state of the matching factor Mm, the risk factors were accurately divided into the first gradient and the second gradient, ensuring that the model could dynamically adjust its evaluation weight when facing different risk factors. This double-layer gradient marking mechanism is particularly suitable for capturing complex risk factor associations, making the model more sensitive and accurate in personalized analysis, and avoiding evaluation errors caused by improper association processing in traditional methods.

[0039] By integrating the complication manifestation set M_Fi with the patient's physical sign data set P, the patient's delirium risk Pd is accurately assessed using a nonlinear calculation formula, significantly improving the sensitivity and timeliness of delirium risk prediction. Different from the traditional single risk assessment, the complication manifestation set M_Fi is more targeted by screening out the key words wp with the first gradient correlation, dynamically matching it with the patient's individual physical sign data P to obtain the specific number of complications Nmatch, and combining the length value MFC of the complication manifestation set M_Fi to accurately calculate the delirium risk Pd. In this process, the model can not only distinguish different risk levels, but also intelligently divide the patient's risk level by matching the delirium risk Pd with the risk assessment threshold Td, and automatically generate corresponding assessment strategies according to different risk levels, including no risk, risk, and delirium warning risk strategies. The unique advantage of this hierarchical strategy evaluation mechanism is that it can not only identify high-risk patients in a timely manner, but also trigger an early warning mechanism when the delirium risk Pd is significantly higher than the threshold. It can achieve accurate and rapid intervention through information tagging and real-time notification of doctors, thereby avoiding the deterioration of the patient's condition and providing a more dynamic and responsive risk management solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 The figure is a schematic diagram of the steps of the method for constructing a risk prediction model for post-stroke delirium of the present invention. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0042] Example 1: The present invention provides a method for constructing a risk prediction model for post-stroke delirium. Figure 1 , including the following steps:

[0043] S1, by collecting cases of risk factors for post-stroke delirium in stroke patients, including control studies, cohort studies, cross-sectional studies, general information of patients, medical history, symptoms after admission, blood indicators and scores of medical staff, integrating documents and performing big data retrieval to find relevant words related to risk factors for post-stroke delirium in relevant patients, and then extracting word occurrence frequency and related sequence information to form the patient's risk factor data set D;

[0044] S2, by performing mutual correlation analysis on the risk factor data set D, forming risk factor data Fi marked with correlation relationships, and reorganizing the risk factor data Fi to obtain a risk factor correlation matrix A;

[0045] S3, traverse and calculate each risk factor data Fi of the risk factor association matrix A, obtain the matching factor Mm between the risk factor data Fi in the association relationship, match it with the preset matching degree screening threshold PZ, and output the gradient marking state result of the matching factor Mm of each risk factor data Fi;

[0046] S4, screening according to the gradient marking state result of the matching factor Mm of each risk factor data Fi, and outputting the complication manifestation set M_Fi of each risk factor data Fi;

[0047] S5, summarize the complication manifestation set M_Fi, and calculate the patient's delirium risk Pd by using a nonlinear calculation formula;

[0048] S6, obtaining the delirium risk assessment strategy for the current patient by matching the obtained delirium risk Pd with the preset risk assessment threshold Td.

[0049] In this embodiment, through steps S1 to S6, the individualized data of the patient is fully utilized, the accuracy of risk prediction is improved, the shortcomings caused by the one-size-fits-all problem in the traditional model are solved, and the whole process from data collection, association analysis to risk assessment is realized. First, by integrating different types of research and individual data of patients, a comprehensive risk factor data set D is established, covering key information such as general information of patients, past medical history, and symptoms after admission. This multi-dimensional data collection method avoids the neglect of individual characteristics of patients and solves the defect that individual differences cannot be accurately adapted in existing models. Then, by calculating the risk factor association matrix A and the matching factor Mm, it is ensured that the association between different symptoms and risk factors is accurately captured and analyzed, especially by screening out the gradient marking state, further enhancing the prediction flexibility and accuracy of the model. The complication performance set M_Fi is matched with the patient data, and the personalized delirium risk Pd is calculated. By comparing with the preset risk assessment threshold Td, a personalized risk assessment strategy is generated. This process ensures that the risk assessment of different patients is more targeted and avoids the misjudgment and delay caused by ignoring individual differences in the traditional model.

[0050] Example 2: This example is an explanation of Example 1. Figure 1 , specifically: S1 includes S11, S12 and S13;

[0051] S11, select stroke patients who meet the inclusion criteria in the hospital's neurology and neurosurgery departments as research subjects, and collect research information on risk factors for post-stroke delirium in stroke patients, including control studies, cohort studies and cross-sectional studies. Combined with case information, general information, past medical history, symptoms after admission, blood indicators and medical staff scores are collected to form the initial patient data set Draw={P1, P2, ..., Pn} for preliminary retrieval, where P1 to Pn represent case patient 1 to case patient n.

[0052] S12, extracting key words w related to delirium risk from the initial patient data set Draw, and counting the frequency of occurrence of the key words w, calculating the frequency fi of the key words w in the case patient data through big data processing technology, and marking the sequence information S of the related key words, the position and order of the key words in the patient's case information, and forming a key word frequency set Q={f1, f2, ..., fi};

[0053] The frequency fi is obtained by the following calculation formula: ; In the formula, fi represents frequency, specifically represents the frequency of the keyword appearing in the initial patient data set Draw, n represents the length of the initial patient data set Draw, specifically represents the total number of case patients included, wm represents the mth keyword wm related to the risk of post-stroke delirium, specifically m represents the subscript information of the keyword w, the mth keyword wm is determined by the combination of wm, and then the mth keyword wm is located by wm, M represents the total number of keywords, specifically obtained by counting the total number of preset search keywords, Pj represents the jth case patient in the initial patient data set Draw, wherein the search keywords specifically include gender, age, previous stroke, previous cognitive impairment, hypertension, hyperlipidemia, diabetes, stroke type, smoking, drinking, atrial fibrillation, ischemic heart disease, heart failure, infection, intubation, medium cell ratio, lymphocyte ratio, fever, urinary incontinence, visual field loss, ADL life ability score, NIHSS score, aphasia and dysphagia keywords.

[0054] S13, extracting sequence information S related to risk factors from the keyword frequency set Q, and simultaneously combining the first to mth keywords and their sequence information S into sequence features, and integrating them into a risk factor data set D;

[0055] The risk factor dataset D is obtained by the following integration method: ; Wherein, Sk represents the kth sequence information of the mth keyword wm.

[0056] In this embodiment, the initial patient data set Draw is generated by integrating the patient data that meets the inclusion criteria, covering a variety of research data such as control studies, cohort studies, and cross-sectional studies, which provides a solid foundation for the subsequent delirium risk keyword extraction. The initial patient data set Draw is analyzed in detail to extract the key words w related to delirium risk, and the frequency of occurrence fi of each keyword and the sequence information S of the keyword in the case information are calculated, which enables the method to not only capture the overall frequency of the keyword, but also identify its specific distribution in different patient cases to form a keyword frequency set Q. Finally, by integrating the key words and their sequence information S into a risk factor data set D, the temporal features and association patterns in the patient data can be effectively retained and refined, so that the risk assessment model can accurately analyze the personalized risk factors of each patient. The special advantage of this process is that it can further refine the risk assessment by introducing the temporal and association features in the patient cases, and enhance the flexibility and dynamic prediction ability of the model.

[0057] Example 3: This example is an explanation of Example 2. Figure 1 , specifically: S2 includes S21 and S22;

[0058] S21, performing association analysis based on sequence information S on the mth keyword wm and the pth keyword wp related to the risk of post-stroke delirium according to the acquired risk factor dataset D, obtaining the association between the appearance patterns of the mth keyword wm and the pth keyword wp at different sequence information S positions, generating an association mark Rmp, and forming risk factor data Fi={(wm, wp, Rmp)} marked with the association relationship;

[0059] The associated mark Rmp is obtained by the following calculation formula: ; Wherein, Rmp represents an association mark, specifically represents the association strength between the mth keyword and the pth keyword, Nm represents the total number of occurrences of the mth keyword, k represents the kth time that the mth keyword wm appears in the total number Nm, Np represents the total number of occurrences of the pth keyword, l represents the lth time that the pth keyword wp appears in the total number Np, represents the average sequence position of the key word wm, represents the average sequence position of the keyword wp, Swmk represents the sequence information of the keyword wm that appears for the kth time, and Swpl represents the sequence information of the keyword wp that appears for the lth time.

[0060] S22, reorganizing the association information of the mth keyword wm and the pth keyword wp according to the generated risk factor data Fi marked with an association relationship, and constructing an association matrix A of associated words, which specifically indicates the mutual association degree between the mth keyword wm and the pth keyword wp;

[0061] The association matrix A is specifically composed in the following way: ; In the formula, each matrix element Rmp in the association matrix A represents the association strength between the keyword wm and the keyword wp, and the diagonal element Rmm=1 represents self-association, wherein the mth keyword wm and the pth keyword wp may be adjacent, or may not be adjacent, and may even be separated by multiple positions.

[0062] The matching factor Mm is obtained by the following calculation formula: ; Wherein, Mm represents the matching factor, specifically the matching factor of the mth keyword wm, amp represents the element in the association matrix A, specifically the association strength between the mth keyword and the pth keyword, and Pp represents the frequency value of the pth keyword in the keyword frequency set Q;

[0063] The matching factor Mm is calculated by traversing the mth row of the association matrix A, that is, calculating the association strength between the key word wm and all other key words wp, and weighting the value of each key word according to these association strengths;

[0064] The gradient marking state result of the matching factor Mm of each risk factor data Fi is obtained by the following matching method: ; When the matching factor Mm≥matching degree screening threshold PZ, the gradient marking state result is obtained as the first gradient marking, and Label=1, indicating that the matching relationship between the mth keyword wm and the pth keyword wp in the risk factor data Fi is marked as 1, specifically indicating that the mth keyword wm and the pth keyword wp are of the first gradient association degree;

[0065] When the matching factor Mm is less than the matching degree screening threshold PZ, the gradient marking state result is obtained as the second gradient marking, and Label=2, indicating that the matching relationship between the mth keyword wm and the pth keyword wp in the risk factor data Fi is marked as 2, specifically indicating that the mth keyword wm and the pth keyword wp are of the second gradient association.

[0066] In the present embodiment, by the S21 and S22 steps in S2, the association analysis and accuracy of the risk assessment model of post-stroke delirium are further improved. First, S21 uses the risk factor data set D to carry out the association analysis based on sequence information to the key words related to post-stroke delirium, not only extracts the association relationship between the key words wm and the key words wp, but also captures the occurrence pattern and mutual relationship of these key words in different positions by calculating the association mark Rmp. This association analysis based on sequence information can identify the temporal association characteristics of risk factors, and the generated association mark Rmp can provide more accurate association information for subsequent model optimization. S22 further reorganizes the risk factor data by the association mark Rmp, constructs the association matrix A between the key words, refines the association strength between each key word, and generates the matching factor Mm of each risk factor data by the traversal calculation of the association matrix. By calculating the gradient marking state of the matching factor Mm, the risk factor is accurately divided into the first gradient and the second gradient, ensuring that the model can dynamically adjust its evaluation weight when facing different risk factors. This double-layer gradient labeling mechanism is particularly suitable for capturing complex risk factor associations, making the model more sensitive and accurate in personalized analysis and avoiding evaluation errors caused by improper association processing in traditional methods.

[0067] Example 4: This example is an explanation of Example 3. Figure 1Specifically: the complication manifestation set M_Fi is integrated with the risk factor data Fi based on the gradient marking status results to obtain the complication manifestation set M_Fi, specifically by screening the pth keyword wp whose mth keyword wm and pth keyword wp have the first gradient correlation, thereby forming the complication manifestation set M_Fi of the risk factor data Fi with only the first gradient correlation.

[0068] The patient's delirium risk Pd is calculated by using a nonlinear calculation formula, wherein the patient obtains general information, past medical history, symptoms after admission, blood indicators and medical staff scoring information through steps S1 to S4 to form the patient's physical sign data set P, and the physical sign data set P is matched with the complication manifestation set M_Fi to obtain the total number of complications Nmatch, and is calculated with the length value MFC of the complication manifestation set M_Fi to obtain the delirium risk Pd;

[0069] Delirium Risk Pd by Obtain the calculation formula.

[0070] The current patient delirium risk assessment strategy is obtained by matching:

[0071] When the delirium risk Pd is less than the risk assessment threshold Td, the delirium risk assessment of the current patient is obtained as no risk strategy;

[0072] When the delirium risk Pd ≥ the risk assessment threshold Td, the delirium risk assessment of the current patient is obtained as a delirium risk strategy, the current patient case information is marked and retained, and a notification is sent to the relevant doctor for consultation;

[0073] When the delirium risk Pd ≥ twice the risk assessment threshold Td, the delirium risk strategy assessment of the current patient is obtained as the delirium warning risk strategy, the current patient case information is marked and retained, and a delirium warning risk strategy notification is sent to relevant doctors for processing.

[0074] In this embodiment, by integrating the complication manifestation set M_Fi with the patient's physical sign data set P, the patient's delirium risk Pd is accurately assessed in combination with a nonlinear calculation formula, which significantly improves the sensitivity and timeliness of delirium risk prediction. Different from the traditional single risk assessment, by screening out the key words wp with the first gradient correlation, the complication manifestation set M_Fi is more targeted, dynamically matched with the patient's individual physical sign data P, and the specific number of complications Nmatch is obtained, and the delirium risk Pd is accurately calculated in combination with the length value MFC of the complication manifestation set M_Fi. In this process, the model can not only distinguish different risk levels, but also intelligently divide the patient's risk level by matching the delirium risk Pd with the risk assessment threshold Td, and automatically generate corresponding assessment strategies according to different risk levels, including no risk, risk, and delirium warning risk strategies. The unique advantage of this hierarchical strategy evaluation mechanism is that it can not only identify high-risk patients in a timely manner, but also trigger an early warning mechanism when the delirium risk Pd is significantly higher than the threshold. It can achieve accurate and rapid intervention through information tagging and real-time notification of doctors, thereby avoiding the deterioration of the patient's condition and providing a more dynamic and responsive risk management solution.

[0075] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a risk prediction model for post-stroke delirium, characterized in that: The following steps are involved: S1, by collecting cases of risk factors for post-stroke delirium in stroke patients, including control studies, cohort studies, cross-sectional studies, general information of patients, medical history, symptoms after admission, blood indicators and scores of medical staff, integrating documents and performing big data retrieval to find relevant words related to risk factors for post-stroke delirium in relevant patients, and then extracting word occurrence frequency and related sequence information to form the patient's risk factor data set D; S2, by performing mutual correlation analysis on the risk factor data set D, forming risk factor data Fi marked with correlation relationships, and reorganizing the risk factor data Fi to obtain a risk factor correlation matrix A; S3, traverse and calculate each risk factor data Fi of the risk factor association matrix A, obtain the matching factor Mm between the risk factor data Fi in the association relationship, match it with the preset matching degree screening threshold PZ, and output the gradient marking state result of the matching factor Mm of each risk factor data Fi; S4, screening according to the gradient marking state result of the matching factor Mm of each risk factor data Fi, and outputting the complication manifestation set M_Fi of each risk factor data Fi; S5, summarize the complication manifestation set M_Fi, and calculate the patient's delirium risk Pd by using a nonlinear calculation formula; S6, obtaining the delirium risk assessment strategy for the current patient by matching the obtained delirium risk Pd with the preset risk assessment threshold Td.

2. The method for constructing a risk prediction model for post-stroke delirium according to claim 1, characterized in that: Said S1 includes S11, S12 and S13; S11, select stroke patients who meet the inclusion criteria in the hospital's neurology and neurosurgery departments as research subjects, and collect research information on risk factors for post-stroke delirium in stroke patients, including control studies, cohort studies and cross-sectional studies. Combined with case information, general information, past medical history, symptoms after admission, blood indicators and medical staff scores are collected to form the initial patient data set Draw={P1, P2, ..., Pn} for preliminary retrieval, where P1 to Pn represent case patient 1 to case patient n.

3. The method for constructing a risk prediction model for post-stroke delirium according to claim 2, characterized in that: S12, extracting key words w related to delirium risk from the initial patient data set Draw, and counting the frequency of occurrence of the key words w, calculating the frequency fi of the key words w in the case patient data through big data processing technology, and marking the sequence information S of the related key words, the position and order of the key words in the patient's case information, and forming a key word frequency set Q={f1, f2, ..., fi}; The frequency fi is obtained by the following calculation formula: ; In the formula, fi represents frequency, specifically represents the frequency of the keyword appearing in the initial patient data set Draw, n represents the length of the initial patient data set Draw, specifically represents the total number of case patients included, wm represents the mth keyword wm related to the risk of post-stroke delirium, specifically m represents the subscript information of the keyword w, the mth keyword wm is determined by the combination of wm, and then the mth keyword wm is located through wm, M represents the total number of keywords, specifically obtained by counting the total number of preset retrieval keywords, and Pj represents the jth case patient in the initial patient data set Draw.

4. The method for constructing a risk prediction model for post-stroke delirium according to claim 3, characterized in that: S13， Extract sequence information S related to risk factors from the keyword frequency set Q, and simultaneously combine the 1st to mth keywords and their sequence information S into sequence features, and integrate them into the risk factor data set D; The risk factor dataset D is obtained by the following integration method: ; Wherein, Sk represents the kth sequence information of the mth keyword wm.

5. The method for constructing a risk prediction model for post-stroke delirium according to claim 1, characterized in that: The S2 includes S21 and S22; S21, performing association analysis based on sequence information S on the mth keyword wm and the pth keyword wp related to the risk of post-stroke delirium according to the acquired risk factor dataset D, obtaining the association between the appearance patterns of the mth keyword wm and the pth keyword wp at different sequence information S positions, generating an association mark Rmp, and forming risk factor data Fi={(wm, wp, Rmp)} marked with the association relationship; The association mark Rmp is obtained by the following calculation formula: ; Wherein, Rmp represents an association mark, specifically represents the association strength between the mth keyword and the pth keyword, Nm represents the total number of occurrences of the mth keyword, k represents the kth time that the mth keyword wm appears in the total number Nm, Np represents the total number of occurrences of the pth keyword, l represents the lth time that the pth keyword wp appears in the total number Np, represents the average sequence position of the key word wm, represents the average sequence position of the keyword wp, Swmk represents the sequence information of the keyword wm that appears for the kth time, and Swpl represents the sequence information of the keyword wp that appears for the lth time.

6. The method for constructing a risk prediction model for post-stroke delirium according to claim 5, characterized in that: S22, reorganizing the association information of the mth keyword wm and the pth keyword wp according to the generated risk factor data Fi marked with an association relationship, and constructing an association matrix A of associated words, which specifically indicates the mutual association degree between the mth keyword wm and the pth keyword wp; The association matrix A is specifically composed in the following manner: ; In the formula, each matrix element Rmp in the association matrix A represents the association strength between keyword wm and keyword wp, and the diagonal element Rmm=1 represents self-association.

7. The method for constructing a risk prediction model for post-stroke delirium according to claim 1, characterized in that: The matching factor Mm is obtained by the following calculation formula: ; Wherein, Mm represents the matching factor, specifically the matching factor of the mth keyword wm, amp represents the element in the association matrix A, specifically the association strength between the mth keyword and the pth keyword, and Pp represents the frequency value of the pth keyword in the keyword frequency set Q; The gradient marking state result of the matching factor Mm of each risk factor data Fi is obtained by the following matching method: ; When the matching factor Mm≥matching degree screening threshold PZ, the gradient marking state result is obtained as the first gradient marking, and Label=1, indicating that the matching relationship between the mth keyword wm and the pth keyword wp in the risk factor data Fi is marked as 1, specifically indicating that the mth keyword wm and the pth keyword wp are of the first gradient association degree; When the matching factor Mm is less than the matching degree screening threshold PZ, the gradient marking state result is obtained as the second gradient marking, and Label=2, indicating that the matching relationship between the mth keyword wm and the pth keyword wp in the risk factor data Fi is marked as 2, specifically indicating that the mth keyword wm and the pth keyword wp are of the second gradient association.

8. The method for constructing a risk prediction model for post-stroke delirium according to claim 7, characterized in that: The complication manifestation set M_Fi is obtained by integrating the risk factor data Fi based on the gradient marking status results, specifically by screening the pth keyword wp whose mth keyword wm and pth keyword wp have a correlation of the first gradient, thereby forming the complication manifestation set M_Fi of the risk factor data Fi with only the correlation of the first gradient.

9. The method for constructing a risk prediction model for post-stroke delirium according to claim 8, characterized in that: The patient's delirium risk Pd is calculated by using a nonlinear calculation formula, wherein the patient obtains general information, past medical history, symptoms after admission, blood indicators and medical staff scoring information through steps S1 to S4 to form the patient's physical sign data set P, and the physical sign data set P is matched with the complication manifestation set M_Fi to obtain the total number of complications Nmatch, and is calculated with the length value MFC of the complication manifestation set M_Fi to obtain the delirium risk Pd; The delirium risk Pd is determined by Obtain the calculation formula.

10. The method for constructing a risk prediction model for post-stroke delirium according to claim 1, characterized in that: The current patient's delirium risk assessment strategy is obtained by matching: When the delirium risk Pd is less than the risk assessment threshold Td, the delirium risk assessment of the current patient is obtained as no risk strategy; When the delirium risk Pd ≥ the risk assessment threshold Td, the delirium risk assessment of the current patient is obtained as a delirium risk strategy, the current patient case information is marked and retained, and a notification is sent to the relevant doctor for consultation; When the delirium risk Pd ≥ twice the risk assessment threshold Td, the delirium risk strategy assessment of the current patient is obtained as the delirium warning risk strategy, the current patient case information is marked and retained, and a delirium warning risk strategy notification is sent to relevant doctors for processing.

Images