A method for constructing a liver cancer patient hypoglycemia risk prediction model

Abstract

The application discloses a liver cancer patient low blood sugar risk prediction model construction method, relates to the medical risk prediction technical field, and has the technical scheme as follows: based on the 6S pyramid model of evidence-based nursing, adopting the PIPOST mode to determine the core elements, searching the related literatures in the preset database and performing quality evaluation, combining the expert score to determine the final independent risk factors, obtaining the modeling group and the verification group sample data by using the sampling method, and constructing the risk prediction model through the Logistic regression analysis; the model performance is verified until the preset standard is met. The application systematically searches the global multiple types of authoritative data sources, screens the independent risk factors and constructs the model according to the strict sample size calculation, accurately quantifies the probability of the liver cancer patient to have low blood sugar, avoids the delay of intervention due to the fact that the low blood sugar symptoms are covered by the liver cancer systemic performance, reduces the incidence of low blood sugar of the liver cancer patient and the adverse consequences such as the irreversible damage caused by low blood sugar, and significantly improves the prognosis of the patient.

Description

Technical Field

[0001] This invention relates to the field of medical risk prediction technology, and more specifically, to a method for constructing a hypoglycemia risk prediction model for liver cancer patients. Background Technology

[0002] The liver, as the core of metabolism for the three major nutrients—carbohydrates, fats, and proteins—is crucial for maintaining stable blood glucose levels. Liver cancer patients, due to impaired liver function, are prone to glucose metabolism disorders. In advanced stages, the large-scale replacement of normal liver tissue by tumor tissue leads to the body's inability to meet the glucose demands of various tissues, thus inducing hypoglycemia. Hypoglycemia is a common and easily overlooked clinical manifestation in the end-stage of liver cancer, with an incidence rate as high as 4% to 27%. Its symptoms are similar to hepatic encephalopathy and are easily masked by the systemic manifestations of liver cancer. If not intervened in time, it can cause dizziness, altered consciousness, or even coma, leading to irreversible damage to vital organs such as the heart and brain, severely impacting the patient's prognosis.

[0003] Currently, research on risk factors for hypoglycemia exhibits significant heterogeneity, with differing conclusions arising from varying study populations, methodologies, and included variables. For instance, some studies indicate that tumor burden and weakened hepatic insulin inactivation are associated with hypoglycemia in patients with liver cancer, while others suggest hospital stay and prothrombin time as relevant risk factors. Regarding risk prediction models, existing technologies primarily focus on patients with type 2 diabetes. For example, hypoglycemia risk prediction models for diabetic patients constructed through prospective surveys and retrospective analyses have seen some clinical application. However, no specific prediction models for patients with liver cancer and concurrent hypoglycemia have been reported.

[0004] The following problems still exist in the application of existing technologies for predicting the risk of hypoglycemia in liver cancer patients: (1) There is a lack of scientific and accurate risk assessment tools for hypoglycemia in liver cancer patients. Existing models are mostly applicable to patients with type 2 diabetes, which cannot match the pathophysiological characteristics and pathogenesis of hypoglycemia in liver cancer patients, and are difficult to meet the needs of clinical screening; (2) The risk factors for hypoglycemia in liver cancer patients are highly heterogeneous. Existing studies have not formed a unified and authoritative screening standard, which makes it impossible to accurately identify high-risk groups; (3) The symptoms of hypoglycemia in liver cancer patients are easily masked. There is a lack of effective predictive risk management methods in clinical practice, making it difficult to intervene in advance to reduce the incidence of hypoglycemia and adverse consequences.

[0005] Therefore, the present invention aims to provide a method for constructing a hypoglycemia risk prediction model for liver cancer patients to solve the above-mentioned problems. Summary of the Invention

[0006] The purpose of this invention is to provide a method for constructing a hypoglycemia risk prediction model for liver cancer patients. Based on the evidence-based nursing 6S pyramid model and PIPOST model, this invention systematically searches authoritative data sources of various types worldwide, and through rigorous sample size calculations, screens independent risk factors and constructs a model. This model can accurately quantify the probability of hypoglycemia in liver cancer patients, avoid delaying intervention due to hypoglycemia symptoms being masked by the systemic manifestations of liver cancer, effectively reduce the incidence of hypoglycemia in liver cancer patients and the adverse consequences of severe hypoglycemia such as dizziness, altered consciousness, and irreversible damage to the heart and brain, and significantly improve patient prognosis.

[0007] The above-mentioned technical objective of the present invention is achieved through the following technical solution: a method for constructing a hypoglycemia risk prediction model for liver cancer patients, comprising the following steps:

[0008] S1. Establishing the research question and retrieval framework: Based on the 6S pyramid model of evidence-based nursing, the PIPOST model is used to clarify the core elements. The core elements include: the research population is liver cancer patients with hypoglycemia; the intervention measures are clinical management measures to prevent hypoglycemia; the participants are medical staff, patients and their families; the outcome indicators are quality of life, incidence of hypoglycemia, patient mortality and severity of hypoglycemia; the research setting is a designated inpatient department of a hospital; and the evidence types are clinical decision-making, evidence summary, clinical guidelines, systematic reviews and expert consensus.

[0009] S2. Literature retrieval and screening: Based on the retrieval framework, relevant literature in the preset database is retrieved, and literature is screened according to the preset inclusion and exclusion criteria to obtain target literature that meets the requirements.

[0010] S3. Literature quality assessment and evidence extraction: At least two independent reviewers will use the JBI Centre for Evidence-Based Healthcare assessment tool from Australia to assess the quality of the target literature. Once the assessments are consistent, risk factors related to hypoglycemia in liver cancer patients will be extracted from the literature to form an initial list of risk factors.

[0011] S4. Expert consultation to verify risk factors: Select experts with preset qualifications and use the Delphi expert consultation method to conduct at least two rounds of consultation on the initial risk factor list. Combine expert scores and opinions to adjust the risk factors and determine the final independent risk factors.

[0012] S5. Improve data collection tools through preliminary investigation: Design a questionnaire based on the final independent risk factors, select a predetermined number of liver cancer patients with hypoglycemia for preliminary investigation, and optimize the questionnaire based on the investigation results and clinical expert opinions.

[0013] S6. Sample Collection and Data Preprocessing: Convenience sampling was used to select liver cancer patients as research subjects, who were divided into a modeling group and a validation group. Clinical data and risk factor-related information of the two groups of patients were collected. The data were cleaned, transformed and reviewed to obtain standardized data.

[0014] S7. Construct a risk prediction model: Use SPSS software to perform univariate and multivariate logistic regression analysis on the standardized data of the modeling group, screen out independent risk factors, and construct a risk prediction model formula based on the independent risk factors and their regression coefficients.

[0015] S8. Model Validation and Optimization: Internal validation is performed using data from the modeling group, and external validation is performed using data from the validation group. The model performance is evaluated through ROC curve analysis and Hosmer-Lemeshow goodness-of-fit test. If the model performance does not meet the standards, the risk factors are adjusted and the modeling and validation steps are repeated until the model performance meets the preset standards.

[0016] The present invention is further configured such that the preset database in step S2 specifically includes:

[0017] Clinical decision-making systems: best clinical practice, clinical decision support systems, Joanna Briggs Institute Evidence-Based Healthcare Center database;

[0018] Guideline websites: UK National Institute for Health and Care Excellence (NIH) Guideline Database, International Guideline Collaboration Network, US National Guideline Database, Scottish Inter-School Guideline Database, National Comprehensive Cancer Guideline Database, World Health Organization Guideline Database, Medlive;

[0019] Academic databases: Wanfang Data Resource System, CNKI Full-text Journal Database, Chongqing VIP Journal Database, Cochrane Library, PubMed.

[0020] The present invention is further configured such that the preset inclusion criteria in step S2 are:

[0021] Study participants were ≥18 years old; met the diagnostic criteria for liver cancer in internal medicine; the literature type was clinical decision-making, best practice information, evidence summary, recommended practice, guideline, systematic review, or expert consensus; and the language was Chinese or English.

[0022] The preset exclusion criteria are:

[0023] The document type is a proposal, draft, or conference content; the information is incomplete, the full text cannot be obtained, the document quality is low, or it is a duplicate publication; the publication date is more than 5 years ago.

[0024] The present invention is further configured such that: the experts with preset qualifications in step S4 meet the following criteria: master's degree or above, associate senior professional title or above, and have worked in the fields of liver cancer nursing, blood glucose management, statistics or oncology for more than 10 years; the number of experts is preset to be at least 10; the Delphi expert consultation method uses the Likert 5-point scoring method to score the importance of risk factors, measures the expert's enthusiasm through the questionnaire response rate, describes the expert's authority through the authority coefficient, measures the degree of opinion concentration through the average value of the indicator importance, and measures the degree of opinion coordination through the coefficient of variation and Kendall's concordance coefficient.

[0025] The present invention is further configured such that: the preset number in step S5 is at least 15 cases, and the questionnaire includes a hypoglycemia risk factor collection form for liver cancer patients and a general information questionnaire.

[0026] The present invention is further configured such that: in step S6, the modeling group and the validation group are divided in a 7:3 ratio; the sample size of the modeling group is calculated according to the formula for calculating the sample size of Logistic regression analysis; 5 to 10 patients are included for each risk factor; considering a 10% sample loss rate, the final sample size is determined in combination with the pre-survey incidence of hypoglycemia of no less than 3.85%; the data preprocessing includes using multiple imputation to handle missing values, using the Z-score algorithm to identify and verify outliers, quantifying qualitative data, and standardizing quantitative data.

[0027] The present invention is further configured such that: in step S6, data collection is achieved through a combination of automatic acquisition via the hospital's electronic medical record system and manual entry via questionnaires; investigators undergo uniform training and use consistent questionnaires and judgment criteria; and the data review involves a secondary review of 5% to 10% of patient data, with different personnel conducting the data analysis.

[0028] The present invention is further configured such that: in step S7, the Logistic regression analysis employs a stepwise forward method combined with the maximum likelihood method, and the risk prediction model formula is:

[0029]

[0030] Where P is the probability of hypoglycemia in liver cancer patients (0≤P≤1). For the intercept term, For each independent risk factor, the regression coefficients are... Assign values ​​to each independent risk factor.

[0031] The present invention is further configured such that: the preset standard in step S8 is AUC≥0.75, Hosmer-Lemeshow test P>0.05; the independent risk factors include at least 6 of the following: tumor burden, concurrent cirrhosis, length of hospital stay, prothrombin time, liver glycogen reserve, glucose metabolism enzyme level, insulin use, sulfonylurea use, combined hypoglycemic drug use, and frequency of self-monitoring of blood glucose.

[0032] This invention also provides a system for constructing a hypoglycemia risk prediction model for liver cancer patients, including a data acquisition module, a data preprocessing module, a risk factor screening module, a model construction module, a risk prediction module, a result display module, an intervention suggestion module, and a system management module;

[0033] The data acquisition module is used to collect clinical data and risk factor information of liver cancer patients, including an electronic medical record interface unit that interfaces with the hospital's electronic medical record system and a questionnaire entry unit for manual data entry. The collected data is transmitted to the data preprocessing module after being encrypted.

[0034] The data preprocessing module is used to clean, transform, and audit the collected raw data to generate standardized data. It includes a data cleaning unit, a data transformation unit, and a data auditing unit. The standardized data is transmitted to the model building module and the risk prediction module.

[0035] The risk factor screening module is used to screen independent risk factors through evidence-based analysis and expert consultation. It includes a literature retrieval and analysis unit, a literature quality evaluation unit, an expert consultation unit, and a risk factor identification unit. The screened independent risk factors are transmitted to the model building module and the data acquisition module.

[0036] The model building module is used to build and validate risk prediction models based on independent risk factors. It includes a statistical analysis unit, a model generation unit, a model validation unit, and a model optimization unit. The completed prediction model and model parameters are transmitted to the risk prediction module, and the model performance report is transmitted to the results display module.

[0037] The risk prediction module is used to receive standardized data from new patients, input it into the prediction model to calculate the risk of hypoglycemia, and includes a data receiving unit, a risk calculation unit and a risk grading unit. The risk probability and risk grading results are transmitted to the result display module and the intervention suggestion module.

[0038] The results display module is used to intuitively display the prediction results and model-related information;

[0039] The intervention suggestion module is used to generate personalized intervention suggestions based on the risk level, including a suggestion generation unit and a suggestion output unit;

[0040] The system management module is used for system user management, access control, data backup, and log recording.

[0041] The present invention is further configured such that: the risk grading unit divides risk levels according to risk probability P: low risk (P<0.3), medium risk (0.3≤P<0.7), and high risk (P≥0.7); the suggestion generation unit automatically matches and generates personalized intervention suggestions based on the risk level and the preset intervention suggestion library of risk factors, and the intervention suggestions include adjustment of blood glucose monitoring frequency, evaluation of hypoglycemic drug dosage, nutritional support plan and health education content.

[0042] The present invention is further configured such that: the data transmission adopts the HTTPS encryption protocol; the system management module sets user permission levels, allowing only authorized personnel to access relevant data; the system data is automatically backed up periodically and operation logs are recorded; the result display module supports the generation of visual charts such as risk radar charts, ROC curves, and calibration curves, which can be accessed by medical staff through terminal devices.

[0043] The present invention also provides a device for constructing a hypoglycemia risk prediction model for liver cancer patients, comprising at least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to implement a method for constructing a hypoglycemia risk prediction model for liver cancer patients.

[0044] In summary, the present invention has the following beneficial effects:

[0045] 1. This invention adopts the 6S pyramid model of evidence-based nursing and the PIPOST model to systematically search multiple types of authoritative data sources worldwide, including clinical decision systems, guideline websites, and academic databases. It combines the evaluation tools of the JBI Centre for Evidence-Based Healthcare in Australia to control the quality of the literature, ensuring the comprehensiveness and scientific nature of risk factor screening. At the same time, it introduces the Delphi expert consultation method to integrate the opinions of authoritative experts in clinical nursing, blood glucose management, and oncology, effectively solving the problem of heterogeneity of literature evidence. The selected independent risk factors are highly consistent with clinical practice, laying a solid foundation for the accuracy of the model.

[0046] 2. This invention employs rigorous sample size calculations, utilizes the requirements of Logistic regression analysis and sample attrition rate, and divides the modeling group and validation group into a 7:3 ratio. Independent risk factors are screened and models are constructed through univariate and multivariate Logistic regression analysis. Then, dual validation and optimization are performed through ROC curve analysis and Hosmer-Lemeshow goodness-of-fit test to ensure that the model has good discrimination and calibration. It can accurately quantify the probability of hypoglycemia in liver cancer patients and provide a reliable basis for clinical risk stratification.

[0047] 3. The model and system constructed in this invention can quickly complete risk assessment after the patient is admitted to the hospital, accurately identify high-risk groups, and enable medical staff to take targeted intervention measures in advance, including adjusting the use of hypoglycemic drugs, increasing the frequency of blood glucose monitoring, and supplementing liver glycogen, so as to avoid delaying intervention due to the systemic manifestations of liver cancer masking hypoglycemia symptoms. This effectively reduces the incidence of hypoglycemia in liver cancer patients and the adverse consequences of severe hypoglycemia such as dizziness, impaired consciousness, and irreversible damage to the heart and brain, and significantly improves the prognosis of patients.

[0048] 4. The system in this invention integrates functions such as automatic data collection, standardized preprocessing, automatic risk calculation, and result visualization, realizing fully automated processing from data input to risk assessment and intervention suggestion generation. It eliminates the need for medical staff to manually organize massive amounts of clinical data and perform complex calculations, significantly shortening the risk assessment time. At the same time, through unified questionnaires and judgment standards, it reduces human error, improves the standardization and efficiency of clinical work, and saves medical staff's energy to focus on patient diagnosis and intervention.

[0049] 5. The model provided by this invention can output the specific hypoglycemia risk probability and risk level of patients. At the same time, the system automatically matches a personalized intervention suggestion library based on the risk level and risk factors, providing medical staff with precise intervention directions, including enhanced blood glucose monitoring for high-risk patients and enhanced nutritional support for intermediate-risk patients. This realizes a personalized management model of "risk stratification - precise intervention", which effectively controls the risk of hypoglycemia while avoiding excessive medical intervention and improving patients' treatment comfort and quality of life.

[0050] 6. The model construction method of this invention follows the norms of evidence-based medicine and statistics, with standardized process and clear logic. It can be promoted and applied in different medical institutions. At the same time, the system adopts a modular design, supports the connection with the electronic medical record system of different hospitals, and is suitable for clinical scenarios in multiple departments of the inpatient department. It is not only applicable to the hypoglycemia risk assessment of liver cancer patients during hospitalization, but also provides extended support for scenarios such as community follow-up and rehabilitation management. It has broad clinical application prospects and promotion value.

[0051] 7. This invention provides a unified and scientific tool for assessing the risk of hypoglycemia in liver cancer patients, standardizes the entire process of risk factor collection, risk assessment, and intervention implementation, reduces the subjectivity and randomness of clinical decision-making, and allows the patient data and intervention effects accumulated during the model's operation to feed back into clinical research, providing real-world data support for the optimization and updating of guidelines for the management of liver cancer complications, and promoting the standardization and evidence-based development of liver cancer nursing. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating the steps of a method for constructing a hypoglycemia risk prediction model for liver cancer patients according to Embodiment 1 of the present invention.

[0053] Figure 2 This is a schematic diagram of the research process for constructing a hypoglycemia risk prediction model for liver cancer patients in Embodiment 1 of the present invention;

[0054] Figure 3 This is a flowchart illustrating the process interaction of a hypoglycemia risk prediction model construction system for liver cancer patients in Embodiment 2 of the present invention.

[0055] Figure 4 This is a schematic diagram of the structural principle of a device for constructing a hypoglycemia risk prediction model for liver cancer patients in Embodiment 3 of the present invention. Detailed Implementation

[0056] The following is in conjunction with the appendix Figures 1-4 The present invention will be described in further detail below.

[0057] Example 1: A method for constructing a hypoglycemia risk prediction model for liver cancer patients, comprising the following steps:

[0058] S1. Establishing the research question and retrieval framework: Based on the 6S pyramid model of evidence-based nursing, the PIPOST model is used to clarify the core elements. The core elements include: the research population is liver cancer patients with hypoglycemia; the intervention measures are clinical management measures to prevent hypoglycemia; the participants are medical staff, patients and their families; the outcome indicators are quality of life, incidence of hypoglycemia, patient mortality and severity of hypoglycemia; the research setting is a designated inpatient department of a hospital; and the evidence types are clinical decision-making, evidence summary, clinical guidelines, systematic reviews and expert consensus.

[0059] In this embodiment, the hierarchy of the "6S" pyramid model is as follows:

[0060] System: The highest level, referring to the integrated Clinical Decision Support System (CDSS), which can directly provide evidence-based clinical decision recommendations without requiring users to integrate evidence themselves;

[0061] Summary: An evidence summary (such as evidence summary or clinical pathway) for a specific clinical question. It has been refined and integrated from the original research for easy and rapid application.

[0062] Synopsis: A structured summary of a single original study, containing core information such as research objectives, methods, results, and conclusions, simplifying the reading of original literature;

[0063] Systematic Review: A study that systematically searches, screens, and evaluates the quality of all relevant original studies on a specific topic, and then integrates them quantitatively or qualitatively, resulting in highly reliable evidence.

[0064] Study (original research): refers to original clinical studies such as randomized controlled trials, cohort studies, and case-control studies, which are the basic sources of evidence;

[0065] Synopsis of Unpublished Evidence: The lowest level, including unpublished research reports, conference abstracts, grey literature, etc., which should be used only after careful evaluation of their reliability.

[0066] The PIPOST mode is as follows:

[0067] Study population (P): Liver cancer patients who have experienced hypoglycemia, aged ≥18 years, and meet the diagnostic criteria for liver cancer in Internal Medicine (Ninth Edition);

[0068] Intervention (I): Focus on various clinical management measures to prevent hypoglycemia, including blood glucose monitoring, adjustment of hypoglycemic drugs, and nutritional support;

[0069] Participants (P): Include healthcare professionals, patients, and their families to ensure that the research results are relevant to clinical practice and patient management needs;

[0070] Outcome measures (O): Quality of life, incidence of hypoglycemia, patient mortality, and severity of hypoglycemia were the core evaluation indicators to comprehensively measure the study effect;

[0071] Research Scenario (S): Focusing on 9 departments in Building 1 and 6 departments in Building 3 of the hospital's inpatient department, consistent with the preliminary investigation scenario to ensure data continuity;

[0072] Evidence type (T): limited to clinical decision-making, evidence summaries, clinical guidelines, systematic reviews, and expert consensus to ensure the authority and reliability of the included evidence.

[0073] S2. Literature Retrieval and Screening: Based on the retrieval framework, relevant literature is retrieved from the preset database, and literature is screened according to the preset inclusion and exclusion criteria to obtain target literature that meets the requirements.

[0074] The search databases are determined to cover three authoritative data sources to ensure comprehensive literature coverage, as detailed below:

[0075] Clinical decision-making systems: Best Clinical Practice (BMJ), UpToDate (a clinical decision support system), and the Joanna Briggs Institute (JBI) Center for Evidence-Based Healthcare database;

[0076] Guideline websites: National Institute for Health and Care Excellence (NICE) Guidelines Database (UK), Guidelines International Network (GIN), National Guidelines Database (NGC) (USA), Scottish Inter-School Guidelines Network (SICN), National Comprehensive Cancer Network (NCCN), World Health Organization Guidelines Network (WHO), Medlive;

[0077] Academic databases: Wanfang Data Resource System, CNKI Full-text Journal Database, Chongqing VIP Journal Database, Cochrane Library, PubMed.

[0078] Set search keywords to ensure a comprehensive and complete search, as shown in Table 1:

[0079] Table 1 Search term list

[0080]

[0081] Limited search period: Search for literature published between the establishment of each database and December 31, 2023, to ensure the timeliness of evidence.

[0082] Perform literature screening:

[0083] Inclusion criteria: Strictly adhered to the four requirements of "age ≥ 18 years", "meets the diagnostic criteria for liver cancer", "the literature type is high-quality evidence such as clinical decision-making and guidelines", and "language is Chinese / English";

[0084] Exclusion criteria: Exclude proposals, drafts, conference proceedings, documents with incomplete information, documents that cannot be fully accessed, documents of low quality or duplicate publications, and documents published more than 5 years ago;

[0085] Screening process: The literature is managed using NoteExpress software. First, the literature is initially screened by title and abstract to remove literature that obviously does not meet the standards. Then, the remaining literature is screened in full text to finally determine the target literature that meets the requirements.

[0086] S3. Literature quality assessment and evidence extraction: At least two independent reviewers will use the JBI Centre for Evidence-Based Healthcare assessment tool from Australia to assess the quality of the target literature. Once the assessments are consistent, risk factors related to hypoglycemia in liver cancer patients will be extracted from the literature to form an initial list of risk factors.

[0087] Quality assessment: Two trained independent evaluators used the corresponding assessment tools from the JBI Centre for Evidence-Based Healthcare in Australia to evaluate the authenticity and reliability of the target literature based on its type (e.g., guidelines, systematic reviews). If the two evaluators disagreed, a third senior researcher was invited to discuss and negotiate until a consensus was reached.

[0088] Evidence extraction: Two researchers independently read the literature that passed the quality assessment and extracted risk factors related to hypoglycemia in liver cancer patients according to the structure of "risk factor name, evidence source, quality level, and recommendation". After extraction, the data were cross-checked to ensure the accuracy of the information, and finally an initial list of risk factors was compiled.

[0089] S4. Expert consultation to verify risk factors: Select experts who meet the preset qualifications, and use the Delphi expert consultation method to conduct at least two rounds of consultation on the initial risk factor list. Combine the expert scores and opinions to adjust the risk factors and determine the final independent risk factors.

[0090] Expert selection: Select 10 experts who meet the preset qualifications, requiring them to meet three conditions: "Master's degree or above", "Associate senior professional title or above" and "more than 10 years of work experience in liver cancer nursing, blood glucose management, statistics or oncology", to ensure that the experts have sufficient authority and professionalism.

[0091] Delphi Consultation: Round 1 Consultation: Distribute an initial list of risk factors and a scoring sheet to experts, use the Likert 5-point scoring method (1 = extremely unimportant, 5 = extremely important) to score the importance of each risk factor, and collect experts' opinions on modification and supplementation of risk factors;

[0092] Second round of consultation: The scoring results and opinions of the first round of consultation are summarized and fed back to the experts. The experts are invited to score again. If there are still significant differences in the experts' opinions, the consultation process is repeated until the experts' opinions tend to be consistent.

[0093] Evaluation of consultation effectiveness: Data was analyzed using Excel 2013 and SPSS 26.0 software. Expert enthusiasm was measured by questionnaire response rate, the degree of expert authority was described by the authority coefficient (Cr, the arithmetic mean of the expert judgment coefficient and the familiarity coefficient), the degree of opinion concentration was measured by the average of the indicator importance values, and the degree of opinion harmony was measured by the coefficient of variation (CV) and Kendall's coefficient of concordance (W). This ensured that the consultation results were scientific and reliable, and finally the final independent risk factors were determined.

[0094] S5. Improve data collection tools through preliminary investigation: Design a questionnaire based on the final independent risk factors, select a predetermined number of liver cancer patients with hypoglycemia for preliminary investigation, and optimize the questionnaire based on the investigation results and clinical expert opinions.

[0095] Design of questionnaires: Based on the final independent risk factors, the "Hypoglycemia Risk Factor Collection Form for Liver Cancer Patients" and the "General Information Questionnaire" were designed. The risk factor collection form covers core indicators such as tumor burden, concurrent cirrhosis, and prothrombin time, while the general information questionnaire includes basic information such as age, gender, and length of hospital stay.

[0096] Preliminary survey: Fifteen patients with liver cancer complicated by hypoglycemia were selected for a clinical preliminary survey. The survey was completed by investigators who had received standardized training, following a standardized procedure.

[0097] Tool optimization: Based on the problems found during the pre-survey (such as vague descriptions of some indicators and unreasonable logical order) and feedback from clinical experts, the questionnaire items were modified and improved to ensure that the questionnaire has good practicality and operability.

[0098] S6. Sample Collection and Data Preprocessing: Convenience sampling was used to select liver cancer patients as research subjects, who were divided into a modeling group and a validation group. Clinical data and risk factor information of the two groups of patients were collected. The data were cleaned, transformed and reviewed to obtain standardized data.

[0099] Study subjects were selected using convenience sampling. Liver cancer patients admitted to the hospital were selected as the study subjects and divided into a modeling group and a validation group, with the two groups divided in a 7:3 ratio.

[0100] Inclusion criteria: age ≥18 years; meeting the diagnostic criteria for liver cancer in Internal Medicine (9th Edition); patient and family consent to participate in this study;

[0101] Exclusion criteria: Individuals with comorbid mental illness, language disorders, or communication difficulties; patients who refused to participate in the study.

[0102] Determine the sample size: Referring to the formula for calculating the sample size in Logistic regression analysis, 5-10 patients are included for each risk factor. Considering a 10% sample attrition rate and the 3.85% hypoglycemia incidence rate in the preliminary survey, if 10 risk factors are finally identified, the sample size of the modeling group should be at least 10×10×(1+0.1)÷3.85%≈2857 cases. The sample size of the validation group is calculated based on the actual number of samples collected in the modeling group at a ratio of 3:7.

[0103] Data Collection: Data collection for the modeling group took place from August to December 2024, and for the validation group from January to March 2025. Patient general information, clinical diagnoses, laboratory tests, and imaging data were automatically acquired through the hospital's electronic medical record system. Investigators manually entered information such as medication use and frequency of self-monitoring of blood glucose using an optimized questionnaire. All investigators underwent standardized training and adhered to consistent judgment criteria to ensure standardized data collection.

[0104] Data preprocessing:

[0105] Data cleaning: missing values ​​are handled using multiple imputation, outliers are identified and verified using the Z-score algorithm, and duplicate data is removed;

[0106] Data transformation: Qualitative data (e.g., gender: male=1, female=0; combined with cirrhosis: yes=1, no=0) are quantified and assigned values, and quantitative data (e.g., age, prothrombin time) are standardized.

[0107] Data review: Designate a specific person to store the data, extract 5% to 10% of the patient data for secondary review, and have different personnel conduct data analysis to ensure data accuracy and finally generate standardized data.

[0108] S7. Constructing a risk prediction model: Using SPSS software, perform univariate and multivariate logistic regression analysis on the standardized data of the modeling group to screen out independent risk factors. Based on the independent risk factors and their regression coefficients, construct a risk prediction model formula.

[0109] Univariate Logistic Regression Analysis: The standardized data of the modeling group were imported into SPSS 22.0 software. With "whether hypoglycemia occurred" as the dependent variable (occurrence = 1, no occurrence = 0) and each risk factor as the independent variable, univariate logistic regression analysis was performed to screen out potential risk factors with P < 0.05.

[0110] Multivariate Logistic Regression Analysis: Potential risk factors screened out by univariate analysis are incorporated into the multivariate logistic regression model. The stepwise forward method combined with the maximum likelihood method is used to control for the influence of confounding factors and screen out independent risk factors.

[0111] Model Formula Establishment: Based on independent risk factors and their regression coefficients, a model formula for predicting the risk of hypoglycemia in liver cancer patients is constructed.

[0112]

[0113] Where P is the probability of hypoglycemia in liver cancer patients (0≤P≤1). For the intercept term, For each independent risk factor, the regression coefficients are... Assign values ​​to each independent risk factor.

[0114] S8. Model Validation and Optimization: Internal validation is performed using data from the modeling group, and external validation is performed using data from the validation group. The model performance is evaluated through ROC curve analysis and Hosmer-Lemeshow goodness-of-fit test. If the model performance does not meet the standards, the risk factors are adjusted and the modeling and validation steps are repeated until the model performance meets the preset standards.

[0115] Internal validation: Internal validation was performed using data from the modeling set. The area under the receiver operating characteristic (ROC) curve (AUC) was calculated to assess the model's discriminative power. The Hosmer-Lemeshow goodness-of-fit test was used to assess the model's calibration. External validation: External validation was performed using data from the validation set. The above validation steps were repeated to assess the model's generalization ability.

[0116] Model optimization: If the model performance does not meet the preset standard (AUC≥0.75, Hosmer-Lemeshow test P>0.05), re-examine the risk factor screening process, supplement or adjust the risk factors, and repeat the modeling and validation process in steps 7-8 until the model performance meets the standard, and finally form a stable and reliable hypoglycemia risk prediction model for liver cancer patients.

[0117] Example 2: A system for constructing a hypoglycemia risk prediction model for liver cancer patients, including a data acquisition module, a data preprocessing module, a risk factor screening module, a model construction module, a risk prediction module, a result display module, an intervention suggestion module, and a system management module.

[0118] In this embodiment, the data acquisition module includes an electronic medical record interface unit and a questionnaire entry unit. The electronic medical record interface unit establishes a real-time connection with the electronic medical record system through the hospital's standardized data interaction protocol, and automatically captures the general information (age, gender, length of hospital stay, etc.), clinical diagnostic information (which must meet the diagnostic criteria for liver cancer in "Internal Medicine (9th Edition)"), laboratory test data (prothrombin time, blood glucose level, etc.), and imaging examination data (tumor size, number, etc.) of liver cancer patients. Basic data collection can be completed without manual intervention, ensuring that the data is consistent with the clinical diagnosis and treatment records.

[0119] The survey entry unit, based on the final independent risk factors output by the risk factor screening module, provides electronic forms for collecting hypoglycemic risk factors in liver cancer patients and general information surveys. This allows healthcare professionals to input information not covered by the electronic medical record system, such as insulin usage, frequency of self-monitoring blood glucose, and use of combined hypoglycemic agents, via hospital terminal devices. During the data entry process, the system incorporates built-in field validation rules, providing real-time alerts for incorrectly formatted or logically contradictory data to ensure data accuracy.

[0120] The collected raw data is encrypted using AES and then transmitted in real time to the data preprocessing module via the hospital's intranet using the HTTPS protocol. The transmitted data packet includes a data collection timestamp, a data source identifier (electronic medical record / manual entry), and an integrity check code. Upon receiving the data, the data preprocessing module verifies its integrity. If any missing or tampered data is found, it automatically sends a retransmission request to the data collection module.

[0121] In this embodiment, the data preprocessing module includes a data cleaning unit, a data conversion unit, and a data review unit. After receiving the raw data, the data cleaning unit uses multiple imputation to complete missing values ​​(such as some laboratory test results not being recorded). It uses the Z-score algorithm to identify outliers (such as prothrombin time exceeding the normal reference range), marks abnormal data, and generates an abnormal report. At the same time, it removes duplicate data based on the patient's unique identifier to avoid data redundancy.

[0122] The data conversion unit quantifies and assigns values ​​to qualitative data according to preset rules. For example, "compound cirrhosis" is assigned "yes=1, no=0", and "insulin use" is assigned "yes=1, no=0". Quantitative data (such as age and tumor size) are processed using the Z-score standardization formula to convert the data to the same order of magnitude to meet the requirements of subsequent statistical analysis.

[0123] The data auditing unit randomly selects 5% to 10% of the processed data and compares it with the original data to verify the data consistency. If the pass rate is lower than 95%, the data reprocessing process is triggered until the audit meets the standards.

[0124] The standardized data, along with processing logs (including missing value completion records, outlier labeling information, and standardized parameters), is transmitted via encrypted channels to the risk factor screening module (for risk factor validation), the model building module (for modeling and validation), and the risk prediction module (for risk assessment of new patients). Simultaneously, a data processing completion notification is sent to the system management module for log recording.

[0125] In this embodiment, the risk factor screening module includes a literature retrieval and analysis unit, a literature quality evaluation unit, an expert consultation unit, and a risk factor identification unit. The literature retrieval and analysis unit, based on the "6S" pyramid model and PIPOST mode, automatically accesses preset databases (including clinical decision-making systems, guideline websites, and academic databases) and performs searches using Chinese and English keywords such as "liver cancer," "hypoglycemia," and "risk factors" (the search time range is from database inception to December 31, 2023). After the search is completed, the unit automatically performs initial screening of literature according to preset inclusion and exclusion criteria (including high-quality evidence from the past 5 years in Chinese / English, excluding drafts, duplicate publications, etc.) and outputs a preliminary screening literature set.

[0126] The literature quality assessment unit utilizes the assessment tools from the JBI Centre for Evidence-Based Healthcare in Australia. Two authorized evaluators independently assess the quality of the literature online through the system. The system records each evaluator's score and comments. If a disagreement arises (score difference ≥ 2 points), a third evaluator is automatically invited to intervene and negotiate. Finally, a collection of literature that meets the quality standards and an assessment report are output.

[0127] The expert consultation unit sent electronic consultation questionnaires to 10 qualified experts (Master's degree or above, associate senior professional title or above, and more than 10 years of experience in liver cancer nursing / blood glucose management / statistics / oncology). A Likert 5-point scoring method (1 = extremely unimportant, 5 = extremely important) was used to collect expert scores and modification opinions on the initial risk factors. Two rounds of consultations were conducted. The system automatically calculated the questionnaire response rate, authority coefficient (Cr), coefficient of variation (CV), and Kendall's concordance coefficient (W). When CV ≤ 0.25 and W passed the significance test, the expert opinions were considered to be concordant.

[0128] The risk factor identification unit combines the initial risk factors extracted from the literature with the results of expert consultation, removes indicators with importance scores below 3 and coefficients of variation greater than 0.25, retains indicators with high consensus, and forms the final list of independent risk factors (such as tumor burden, cirrhosis, prothrombin time, etc.).

[0129] The final list of independent risk factors is transmitted in encrypted JSON format to the model building module as the core input variable for modeling; it is also transmitted to the data acquisition module to update the survey form fields (adding fields for high-importance risk factors and deleting fields for low-importance risk factors); expert consultation data and literature evaluation reports are transmitted to the system management module for archiving, and the archived data is stored in encrypted form.

[0130] In this embodiment, the model building module includes a statistical analysis unit, a model generation unit, a model validation unit, and a model optimization unit. The statistical analysis unit receives standardized data from the modeling group transmitted by the data preprocessing module, calls the SPSS 22.0 statistical analysis interface, and uses "whether hypoglycemia occurs" as the dependent variable (occurrence = 1, non-occurrence = 0), with the final independent risk factors as independent variables. First, a univariate logistic regression analysis is performed to screen out potential risk factors with P < 0.05. Then, the potential risk factors are included in a multivariate logistic regression analysis, using a stepwise forward method combined with the maximum likelihood method to control for the influence of confounding factors and screen out at least 6 items from tumor burden, cirrhosis, length of hospital stay, prothrombin time, glycogen reserve, glucose metabolism enzyme level, insulin use, sulfonylurea use, combined hypoglycemic drug use, and frequency of self-monitoring of blood glucose. In this embodiment, based on the logistic regression analysis results, the number of independent risk factors with significant regression coefficients among the above 10 independent risk factors is 6-8, so at least 6 items are used as the basic variable set for model building.

[0131] The model generation unit automatically constructs risk prediction model formulas based on independent risk factors and their regression coefficients.

[0132]

[0133] Where P is the probability of hypoglycemia in liver cancer patients (0≤P≤1). For the intercept term, For each independent risk factor, the regression coefficients are... Assign values ​​to each independent risk factor and calculate the weight percentage of each risk factor.

[0134] The model validation unit uses modeling group data for internal validation, calculates the area under the curve (AUC) through ROC curve analysis to evaluate the model's discrimination, and uses the Hosmer-Lemeshow goodness-of-fit test to evaluate the model's calibration. Then, it uses validation group data (divided from the standardized data in a 7:3 ratio) for external validation, repeats the above validation process, and outputs a validation report.

[0135] If the model optimization unit detects that the model performance does not meet the standard (AUC < 0.75 or Hosmer-Lemeshow test P ≤ 0.05), it automatically sends an optimization request to the risk factor screening module to re-screen or adjust the risk factors, and repeat the statistical analysis and model generation steps until the model performance meets the preset standard.

[0136] The validated prediction model and its parameters (intercept, regression coefficient, risk factor weights) are encrypted and transmitted to the risk prediction module for new patient risk calculation; the model performance report (including AUC value, calibration test results, and validation data statistics) is transmitted to the results display module; the raw data and intermediate statistical results during the modeling process are transmitted to the system management module for backup, supporting subsequent traceability and model iteration.

[0137] In this embodiment, the risk prediction module includes a data receiving unit, a risk calculation unit, and a risk grading unit. The data receiving unit receives new standardized patient data transmitted by the data preprocessing module in real time, verifies the data integrity, checks whether all independent risk factor fields are included, and if any fields are missing, automatically sends a supplementary entry reminder to the data acquisition module until the data is complete.

[0138] The risk calculation unit calls the prediction model transmitted by the model building module, substitutes the standardized data of the new patient into the model formula, and automatically calculates the probability P of hypoglycemia. The calculation process runs in the background without manual intervention, ensuring rapid output of results.

[0139] The risk grading unit divides risk levels according to preset rules: P < 0.3 is low risk, 0.3 ≤ P < 0.7 is medium risk, and P ≥ 0.7 is high risk. This generates risk grading results, clarifying the patient's risk level. In this embodiment, the risk probability thresholds of 0.3 and 0.7 are based on relevant industry safety standards and risk assessment specifications regarding acceptable and unacceptable risk levels, determining the critical values ​​for low, medium, and high risk. Extensive simulation experiments and fitting of field measurement data are used to verify the risk consequences under different probability intervals, ensuring that the threshold division matches the actual degree of harm. Integrating the experience of experts in the field, and while ensuring the model's simplicity and feasibility, 0.3 and 0.7 are determined as risk level cutoff thresholds, making the grading results objective, stable, and repeatable.

[0140] The risk probability P and risk grading results are transmitted synchronously through an encrypted channel to the results display module (for medical staff to view) and the intervention suggestion module (for generating intervention suggestions). The transmitted data packet includes the patient's unique identifier and the risk calculation timestamp to ensure that the results are accurately matched with the patient's information and to avoid confusion.

[0141] In this embodiment, the result display module includes a prediction result display unit, a model information display unit, and a visualization unit. The prediction result display unit receives the output data from the risk prediction module and, combined with the patient's basic information synchronously obtained from the data acquisition module, clearly displays the specific values ​​of each risk factor, the probability of hypoglycemia, and the risk level of the patient in tabular form, so that medical staff can quickly grasp the core information.

[0142] The model information display unit shows the core parameters of the currently used prediction model, including AUC value, calibration test results, independent risk factors and their weights, so that medical staff can understand the reliability of the model and provide a reference for clinical decision-making.

[0143] The visualization unit automatically generates risk radar charts (intuitively presenting the contribution of each risk factor to patient risk), ROC curves (showing the model's discriminative power), and calibration curves (showing the model's calibration accuracy). It supports medical staff in zooming in and downloading charts to meet the needs of clinical reporting or case analysis.

[0144] The displayed content is transmitted to medical staff's terminal devices (computers, tablets) through dynamic encryption based on user permissions assigned by the system management module. Only authorized users can decrypt and view the data, and unauthorized users cannot access the relevant data, thus protecting patient privacy.

[0145] In this embodiment, the intervention suggestion module includes a suggestion generation unit and a suggestion output unit. The suggestion generation unit has a built-in intervention suggestion library containing personalized suggestions corresponding to different risk levels and combinations of risk factors. For example, for high-risk patients, the suggestion is to "increase the frequency of blood glucose monitoring to once every 2 hours, assess the need for insulin dose adjustment, and strengthen liver glycogen supplementation"; for intermediate-risk patients, the suggestion is to "check blood glucose 3 times a week and optimize the nutritional support plan"; and for low-risk patients, the suggestion is to "provide routine care and conduct hypoglycemia prevention health education once a month". After receiving the risk level and patient risk factor data from the risk prediction module, the module automatically matches the optimal intervention suggestion.

[0146] The output unit suggests that the matched intervention recommendations be output in text form, allowing medical staff to manually modify and supplement the recommendations based on the actual clinical situation, ensuring the flexibility and practicality of the recommendations.

[0147] The final intervention recommendations and predicted results are displayed simultaneously in the results display module. Medical staff can export assessment reports in PDF format (including predicted results, risk factor analysis, and intervention recommendations). The exported files are automatically encrypted and can only be opened and viewed by authorized users.

[0148] In this embodiment, the system management module includes a user management unit, a data backup unit, and a log recording unit. The user management unit supports administrators in creating, modifying, and deleting user accounts, and assigning permissions according to roles: medical staff can only view patient data and operational risk prediction functions within their department; researchers can access modeling data and model parameters, but cannot modify core configurations; administrators have full access and are responsible for system maintenance. User login and operation history are also recorded for easy permission tracing.

[0149] The data backup unit automatically backs up all data in the system at a fixed time every day, including patient data, model data, literature data, operation logs, etc. The backup data is stored on a designated server in the hospital and supports data recovery by time point to prevent data loss.

[0150] The log recording unit records the operational behavior of each module of the system in real time, including data collection, preprocessing, modeling, risk prediction, and result export. The recorded content includes the operator, operation time, operation content, and data changes, forming an immutable operation log for audit traceability and ensuring the compliant operation of the system.

[0151] This module maintains real-time communication with all other modules, receives operation status notifications and data transmission requests from each module, and sends permission verification results and data backup instructions to each module to ensure the efficient and coordinated operation of all aspects of the system.

[0152] Example 3: A device for constructing a hypoglycemia risk prediction model for liver cancer patients, comprising at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to implement a method for constructing a hypoglycemia risk prediction model for liver cancer patients, the method comprising: establishing a research question and retrieval framework: based on the evidence-based nursing 6S pyramid model, using the PIPOST model to clarify the core elements, the core elements including the research population being liver cancer patients experiencing hypoglycemia, the intervention measures being clinical management measures for preventing hypoglycemia, the participants being medical staff, patients and their families, and the outcome indicators... The criteria for inclusion were: quality of life, incidence of hypoglycemia, patient mortality, and severity of hypoglycemia; the research setting was a designated inpatient department in a hospital; and the evidence types were clinical decision-making, evidence summaries, clinical guidelines, systematic reviews, and expert consensus. Literature retrieval and screening: Based on the retrieval framework, relevant literature was retrieved from pre-defined databases, and literature was screened according to pre-defined inclusion and exclusion criteria to obtain target literature that met the requirements. Literature quality assessment and evidence extraction: At least two independent reviewers used the JBI Centre for Evidence-Based Healthcare assessment tool from Australia to assess the quality of the target literature. After consensus was reached, risk factors related to hypoglycemia in liver cancer patients were extracted from the literature to form an initial risk factor list. Single; Expert consultation to validate risk factors: Select qualified experts and use the Delphi expert consultation method to conduct at least two rounds of consultation on the initial risk factor list. Adjust the risk factors based on expert scores and opinions to determine the final independent risk factors; Pre-survey to improve data collection tools: Design a questionnaire based on the final independent risk factors, select a predetermined number of liver cancer patients with hypoglycemia for a pre-survey, and optimize the questionnaire based on the survey results and clinical expert opinions; Sample collection and data preprocessing: Select liver cancer patients as research subjects using convenience sampling, divided into a modeling group and a validation group. Collect clinical data and risk factor-related information from both groups of patients, and logarithmically... The data was cleaned, transformed, and reviewed to obtain standardized data. A risk prediction model was constructed: SPSS software was used to perform univariate and multivariate logistic regression analysis on the standardized data of the modeling group to screen out independent risk factors. Based on these independent risk factors and their regression coefficients, a risk prediction model formula was constructed. Model validation and optimization: internal validation was performed using data from the modeling group, and external validation was performed using data from the validation group. Model performance was evaluated through ROC curve analysis and Hosmer-Lemeshow goodness-of-fit tests. If the model performance did not meet the standards, the risk factors were adjusted, and the modeling and validation steps were repeated until the model performance met the preset standards.

[0153] This specific embodiment is merely an explanation of the present invention and is not intended to limit the invention. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A method for constructing a hypoglycemia risk prediction model for liver cancer patients, characterized in that: Includes the following steps: S1. Establishing the research question and retrieval framework: Based on the 6S pyramid model of evidence-based nursing, the PIPOST model is used to clarify the core elements. The core elements include: the research population is liver cancer patients with hypoglycemia; the intervention measures are clinical management measures to prevent hypoglycemia; the participants are medical staff, patients and their families; the outcome indicators are quality of life, incidence of hypoglycemia, patient mortality and severity of hypoglycemia; the research setting is a designated inpatient department of a hospital; and the evidence types are clinical decision-making, evidence summary, clinical guidelines, systematic reviews and expert consensus. S2. Literature retrieval and screening: Based on the retrieval framework, relevant literature in the preset database is retrieved, and literature is screened according to the preset inclusion and exclusion criteria to obtain target literature that meets the requirements. S3. Literature quality assessment and evidence extraction: At least two independent reviewers will use the JBI Centre for Evidence-Based Healthcare assessment tool from Australia to assess the quality of the target literature. Once the assessments are consistent, risk factors related to hypoglycemia in liver cancer patients will be extracted from the literature to form an initial list of risk factors. S4. Expert consultation to verify risk factors: Select experts with preset qualifications and use the Delphi expert consultation method to conduct at least two rounds of consultation on the initial risk factor list. Combine expert scores and opinions to adjust the risk factors and determine the final independent risk factors. S5. Improve data collection tools through preliminary investigation: Design a questionnaire based on the final independent risk factors, select a predetermined number of liver cancer patients with hypoglycemia for preliminary investigation, and optimize the questionnaire based on the investigation results and clinical expert opinions. S6. Sample Collection and Data Preprocessing: Convenience sampling was used to select liver cancer patients as research subjects, who were divided into a modeling group and a validation group. Clinical data and risk factor-related information of the two groups of patients were collected. The data were cleaned, transformed and reviewed to obtain standardized data. S7. Construct a risk prediction model: Use SPSS software to perform univariate and multivariate logistic regression analysis on the standardized data of the modeling group, screen out independent risk factors, and construct a risk prediction model formula based on the independent risk factors and their regression coefficients. S8. Model Validation and Optimization: Internal validation is performed using data from the modeling group, and external validation is performed using data from the validation group. The model performance is evaluated through ROC curve analysis and Hosmer-Lemeshow goodness-of-fit test. If the model performance does not meet the standards, the risk factors are adjusted and the modeling and validation steps are repeated until the model performance meets the preset standards.

2. The method for constructing a hypoglycemia risk prediction model for liver cancer patients according to claim 1, characterized in that: The preset database in step S2 specifically includes: Clinical decision-making systems: best clinical practice, clinical decision support systems, Joanna Briggs Institute Evidence-Based Healthcare Center database; Guideline websites: UK National Institute for Health and Care Excellence (NIH) Guideline Database, International Guideline Collaboration Network, US National Guideline Database, Scottish Inter-School Guideline Database, National Comprehensive Cancer Guideline Database, World Health Organization Guideline Database, Medlive; Academic databases: Wanfang Data Resource System, CNKI Full-text Journal Database, Chongqing VIP Journal Database, Cochrane Library, PubMed.

3. The method for constructing a hypoglycemia risk prediction model for liver cancer patients according to claim 1, characterized in that: The preset inclusion criteria in step S2 are: Study participants were ≥18 years old; met the diagnostic criteria for liver cancer in internal medicine; the literature type was clinical decision-making, best practice information, evidence summary, recommended practice, guideline, systematic review, or expert consensus; and the language was Chinese or English. The preset exclusion criteria are: The document type is a proposal, draft, or conference content; the information is incomplete, the full text cannot be obtained, the document quality is low, or it is a duplicate publication; the publication date is more than 5 years ago.

4. The method for constructing a hypoglycemia risk prediction model for liver cancer patients according to claim 1, characterized in that: In step S4, the pre-defined qualifications of the experts are: master's degree or above, associate senior professional title or above, and more than 10 years of experience in liver cancer nursing, blood glucose management, statistics or oncology. The number of experts is pre-defined to be at least 10. The Delphi expert consultation method uses the Likert 5-point scoring method to score the importance of risk factors, measures the experts' enthusiasm through the questionnaire response rate, describes the experts' authority through the authority coefficient, measures the degree of consensus through the average value of the indicator importance, and measures the degree of consensus through the coefficient of variation and Kendall's concordance coefficient.

5. The method for constructing a hypoglycemia risk prediction model for liver cancer patients according to claim 1, characterized in that: The preset number in step S5 is at least 15 cases, and the questionnaire includes a hypoglycemia risk factor collection form for liver cancer patients and a general information questionnaire.

6. The method for constructing a hypoglycemia risk prediction model for liver cancer patients according to claim 1, characterized in that: In step S6, the modeling group and the validation group are divided in a 7:3 ratio. The sample size of the modeling group is calculated according to the formula for calculating the sample size of Logistic regression analysis. 5 to 10 patients are included for each risk factor. Considering a 10% sample dropout rate, the final sample size is determined by combining the pre-survey incidence of hypoglycemia of no less than 3.85%. The data preprocessing includes using multiple imputation to handle missing values, using the Z-score algorithm to identify and verify outliers, quantifying qualitative data, and standardizing quantitative data.

7. The method for constructing a hypoglycemia risk prediction model for liver cancer patients according to claim 1, characterized in that: In step S6, data collection is achieved through a combination of automatic acquisition via the hospital's electronic medical record system and manual entry via questionnaires. Investigators undergo standardized training and use consistent questionnaires and judgment criteria. Data review involves a secondary review of 5% to 10% of patient data, with different personnel conducting the data analysis.

8. The method for constructing a hypoglycemia risk prediction model for liver cancer patients according to claim 1, characterized in that: In step S7, the Logistic regression analysis employs a combination of the stepwise forward method and the maximum likelihood method. The risk prediction model formula is as follows: Where P is the probability of hypoglycemia in liver cancer patients (0≤P≤1). For the intercept term, For each independent risk factor, the regression coefficients are... Assign values ​​to each independent risk factor.

9. The method for constructing a hypoglycemia risk prediction model for liver cancer patients according to claim 1, characterized in that: The preset standard in step S8 is AUC≥0.75 and Hosmer-Lemeshow test P>0.05; the independent risk factors include at least 6 of the following: tumor burden, cirrhosis, length of hospital stay, prothrombin time, liver glycogen reserve, glucose metabolism enzyme level, insulin use, sulfonylurea use, combined hypoglycemic drug use, and frequency of self-monitoring of blood glucose.

10. A system for constructing a hypoglycemia risk prediction model for liver cancer patients, applied to the method for constructing a hypoglycemia risk prediction model for liver cancer patients as described in any one of claims 1-9, characterized in that: It includes a data acquisition module, a data preprocessing module, a risk factor screening module, a model building module, a risk prediction module, a results display module, an intervention suggestion module, and a system management module; The data acquisition module is used to collect clinical data and risk factor information of liver cancer patients, including an electronic medical record interface unit that interfaces with the hospital's electronic medical record system and a questionnaire entry unit for manual data entry. The collected data is transmitted to the data preprocessing module after being encrypted. The data preprocessing module is used to clean, transform, and audit the collected raw data to generate standardized data. It includes a data cleaning unit, a data transformation unit, and a data auditing unit. The standardized data is transmitted to the model building module and the risk prediction module. The risk factor screening module is used to screen independent risk factors through evidence-based analysis and expert consultation. It includes a literature retrieval and analysis unit, a literature quality evaluation unit, an expert consultation unit, and a risk factor identification unit. The screened independent risk factors are transmitted to the model building module and the data acquisition module. The model building module is used to build and validate risk prediction models based on independent risk factors. It includes a statistical analysis unit, a model generation unit, a model validation unit, and a model optimization unit. The completed prediction model and model parameters are transmitted to the risk prediction module, and the model performance report is transmitted to the results display module. The risk prediction module is used to receive standardized data from new patients, input it into the prediction model to calculate the risk of hypoglycemia, and includes a data receiving unit, a risk calculation unit and a risk grading unit. The risk probability and risk grading results are transmitted to the result display module and the intervention suggestion module. The results display module is used to intuitively display the prediction results and model-related information; The intervention suggestion module is used to generate personalized intervention suggestions based on the risk level, including a suggestion generation unit and a suggestion output unit; The system management module is used for system user management, access control, data backup, and log recording.

11. The system for constructing a hypoglycemia risk prediction model for liver cancer patients according to claim 10, characterized in that: The risk grading unit classifies risk levels according to risk probability P: low risk (P < 0.3), medium risk (0.3 ≤ P < 0.7), and high risk (P ≥ 0.7). The suggestion generation unit automatically matches and generates personalized intervention suggestions based on the risk level and a preset intervention suggestion library of risk factors. The intervention suggestions include adjustment of blood glucose monitoring frequency, assessment of hypoglycemic drug dosage, nutritional support plan, and health education content.

12. The system for constructing a hypoglycemia risk prediction model for liver cancer patients according to claim 10, characterized in that: The data transmission uses the HTTPS encryption protocol. The system management module sets user permission levels, allowing only authorized personnel to access relevant data. It also automatically backs up system data and records operation logs regularly. The results display module supports the generation of visual charts such as risk radar charts, ROC curves, and calibration curves, which can be accessed by medical staff through terminal devices.

13. A device for constructing a hypoglycemia risk prediction model for liver cancer patients, characterized in that: The device includes at least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor to implement a method for constructing a hypoglycemia risk prediction model for liver cancer patients according to any one of claims 1-9.

Images