Prediction system and method for full stomach risk of patient in selective painless gastrointestinal endoscopy

Abstract

The invention discloses a system for predicting the full stomach risk of a patient in selective painless gastrointestinal endoscopy. The system comprises a sample data acquisition module, a sample data processing module, a feature selection module, a model construction module and a model verification module, the sample data acquisition module is used for screening to obtain candidate predictive factors related to full stomach according to clinical data of a patient, and calculating a sample size; the sample data processing module preprocesses the data; the feature selection module is used for performing feature selection on the sample data in the training set to construct an optimal feature set; the model construction module is used for constructing an individualized full stomach risk prediction function model before painless gastrointestinal endoscopy; and the model verification module evaluates and verifies the full stomach prediction model by adopting the test set. By quantifying the probability values of various predictive factors, the gastric contents are objectively scored, and the system is a visual and continuous evaluation system, and is lower in evaluation cost, higher in efficiency and easier to popularize and use.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a system for predicting the risk of stomach fullness in patients undergoing elective painless gastrointestinal endoscopy. Background Art

[0002] Procedural sedation can weaken or eliminate the protective choking and swallowing reflexes of patients undergoing painless gastrointestinal endoscopy, weaken gastrointestinal motility and lower esophageal sphincter function, and reduce the difference between the sphincter pressure and the intragastric pressure, leading to reflux of gastric contents and pulmonary aspiration. Reflux and aspiration of gastric contents into the respiratory tract is the most common complication during the perioperative period, and can even cause death in severe cases. The assessment of fullness before digestive endoscopy has always been a major concern in the academic community. In order to reduce residual gastric contents, patients undergoing painless gastrointestinal endoscopy should follow the fasting and drinking management, but some patients with delayed gastric emptying (about 12%) will still have a full stomach. Therefore, identifying a full stomach before sedation is the key to avoiding aspiration, which can provide clues for the fasting and drinking regimen and anesthesia strategy selection for patients undergoing painless gastrointestinal endoscopy. Currently, there are three ways to evaluate the stomach by ultrasound before endoscopy: 1. Qualitative evaluation under ultrasound, when the patient has liquid or solid gastric contents in both the right lateral and supine positions, it indicates a full stomach, corresponding to a high risk of aspiration; 2. Gastric ultrasound examination with the patient in the supine or right lateral position, and the cross-sectional area (CSA) of the gastric antrum measured under ultrasound reaches 340 mm 2 A value of 1:1 or above can indicate that the patient is in a state of full stomach, which may lead to reflux and aspiration of gastric contents. CSA in the right lateral position has a higher sensitivity in predicting full stomach. ③ A linear regression model based on gastric ultrasound (the cranio-caudal diameter, anteroposterior diameter and CSA of the gastric antrum cross section) is used to assess full stomach. Among them, CSA data in two body positions are widely recognized. They are based on measuring the CSA of adults in the semi-supine position and the right lateral position, respectively. The regression model is used to calculate the volume of gastric contents, and then judge full stomach and assess the risk of aspiration.

[0003] With the development of anesthesia medicine, some anesthesiologists have mastered ultrasound visualization technology, which is also widely used in anesthesia inside and outside the operating room. The working mode of anesthesiologists to evaluate the risk of full stomach in patients undergoing elective painless gastrointestinal endoscopy has gradually transitioned from clinical experience judgment to gastric ultrasound evaluation and a combination of the two methods. However, both methods have defects. The disadvantages of clinical experience judgment are: 1. It is too subjective to judge whether the patient is full of stomach based on the clinical experience of the anesthesiologist, and the reproducibility of the judgment results of different physicians is not high enough; 2. There are significant individual differences between patients with diseases related to delayed gastric emptying, resulting in many confounding factors, so the clinical subjective judgment method has large errors. Clinically, 55% of patients whose gastric juice volume is judged as full under direct gastroscopy are misjudged as empty by clinical experience. The disadvantages of gastric ultrasound evaluation are: 1. Ultrasound equipment has not yet been popularized in anesthesia sites outside the operating room: Due to insufficient investment in medical equipment, ultrasound equipment is not adequately equipped in the digestive endoscopy comfort medical center, making it difficult to carry out gastric ultrasound technology. 2. The proportion of anesthesiologists who are proficient in gastric ultrasound technology is not high: Although transabdominal gastric ultrasound technology does not require detailed evaluation like ultrasound physicians, the qualitative and quantitative evaluation of gastric content by gastric ultrasound still requires anesthesiologists to master relevant ultrasound technology and anatomical basis, and has certain requirements for operation skills. The proportion of anesthesiologists who are proficient in gastric ultrasound technology in China is not high, so it is impossible to carry out gastric content evaluation under preoperative gastric ultrasound on a large scale. 3. Patient flow pressure: With the increasing demand of the people for comfortable medical care, the number of painless gastroenteroscopy diagnosis and treatment has increased rapidly in recent years. The gastric ultrasound operation time of skilled anesthesiologists is generally within a few minutes. The daily work situation of the digestive endoscopy center is that the number of patients for painless gastroenteroscopy diagnosis and treatment is large, the turnover is fast, and the anesthesiologists with highly saturated workload cannot guarantee sufficient time to perform gastric ultrasound evaluation on each patient before anesthesia. In particular, when patients suffer from hunger, fatigue and other discomforts due to long-term fasting, the efficiency of endoscopic examination is particularly important. 4. Increased economic burden on patients: Gastric ultrasound evaluation has not yet been included in the list of routine diagnosis and treatment items. This additional examination item brings a certain degree of economic burden to patients.

[0004] The nomogram prediction system has been widely used as one of the clinical statistical methods, which not only has the advantage of visualization, but also helps to calculate the probability of risk factors. The invention with the patent number CN115274115A mentions a method for constructing a nomogram prediction model for prolonging the mechanical ventilation time of patients after cardiac and vascular surgery. The invention calculates the probability of the occurrence of endpoint events by constructing a nomogram prediction system, which can help medical staff make decisions and the perioperative management of patients undergoing cardiac and vascular surgery, and can also help the rational allocation of medical resources after surgery in the cardiac and vascular surgery intensive care unit (CSICU), and will provide managers, doctors and nurses with estimates of hospital stay. However, when selecting potential risk factors, the invention uses a univariate analysis method to analyze two groups of clinical groups, and screens out some statistically significant factors. The screened factors are basically clinical feature data, which is more suitable for postoperative management fields such as cardiac and vascular surgery. For painless gastrointestinal endoscopy with unclear clinical data characteristics, this method is no longer applicable. Therefore, there is currently a lack of an accurate, fast, cheap, and easy-to-popular method for predicting the risk of fullness. According to searches on the official website of the State Intellectual Property Office, China National Knowledge Infrastructure, Pubmed, and Web of Science, no relevant patents or Chinese or English literature reports have been found so far in the field of prediction tools for the risk of fullness in patients undergoing elective painless gastrointestinal endoscopy. Summary of the invention

[0005] The technical problem to be solved by the present invention is as follows: In view of the many problems existing in the existing methods for predicting and evaluating stomach fullness, such as high personnel costs, large errors, and heavy economic burden, the present invention provides a stomach fullness risk prediction system for patients undergoing elective painless gastrointestinal endoscopy. By quantifying the probability values ​​of various prediction factors, the stomach contents are objectively scored. It is a visual and continuous evaluation system with lower evaluation costs, higher efficiency, and easier promotion and use.

[0006] The present invention adopts the following technical solutions to solve the above technical problems:

[0007] The present invention proposes a stomach fullness risk prediction system for patients undergoing elective painless gastrointestinal endoscopy, the stomach fullness risk assessment system comprising a sample data collection module, a sample data processing module, a feature selection module, a model construction module and a model verification module;

[0008] The sample data collection module is used to screen candidate predictors related to fullness based on the patient's clinical data, the candidate predictors including the patient's demographic data, lifestyle variables, laboratory data, preoperative management data and clinical characteristic data; the sample size is calculated based on the ratio of the number of individuals with outcome events to the number of candidate predictor variables;

[0009] The sample data processing module is used to add digital labels to the candidate prediction factors output by the sample data acquisition module and the satiety result indicators corresponding to the patient, generate a sample data set, and divide the sample data set into a training set and a test set;

[0010] The feature selection module is used to perform feature selection on the sample data in the training set. Specifically, the variables related to the satiety outcome index are screened out using the univariate analysis method, and the variables with significant univariate analysis are subjected to multivariate logistic regression. The variables with significant multivariate logistic regression are then included in the forward stepwise regression to screen out the predictors of satiety. Finally, the variation inflation factor is used to evaluate the collinearity between the predictors, and after excluding the predictors with significant collinearity, the optimal feature set is constructed.

[0011] The model building module builds an individualized full stomach risk prediction function model before painless gastroenteroscopy: logit P = B1×P1+B2×P2+B3×P3+…Bn×Pn+constant, and uses the best feature set to train the individualized full stomach risk prediction function model to obtain a full stomach prediction model;

[0012] The model validation module uses a test set to evaluate and validate the satiety prediction model.

[0013] Furthermore, the demographic characteristics include the patient's age, body mass index and gender; lifestyle variables include the patient's smoking history and drinking history; laboratory data include hemoglobin; preoperative management data include preoperative fasting time and preoperative diet type; clinical characteristics include history of mental illness, history of hypothyroidism, history of diabetes, history of cirrhosis, history of chronic renal dysfunction, history of cerebral infarction, sequelae of cerebral infarction, history of PPI or H2 receptor antagonist medication, history of prokinetic drug use and type of endoscopy to be examined.

[0014] Further, the sample data collection module includes a sample size calculation component;

[0015] The sample size calculation component calculates the sample size for each variable based on the empirical rule of multivariate logic, and the calculation formula is as follows:

[0016]

[0017] Furthermore, the sample data processing module includes a sample division component, a sample deletion component and a sample interpolation component;

[0018] The sample partitioning component partitions the sample data set into a training set and a test set in a ratio of 7:3;

[0019] The sample deletion component is used to delete variables with more than 25% missing values ​​in the training set;

[0020] The sample interpolation component interpolates variables with missing values ​​not exceeding 25% to generate a baseline data table; wherein continuous variables are interpolated using the KNN method, and categorical variables are interpolated using the majority method.

[0021] Further, the feature selection module includes a feature screening component, a feature analysis component and a feature construction component;

[0022] The feature screening component was screened using univariate analysis to screen out variables with a P value of <0.05 in the training set, which were defined as variables associated with the outcome; among them, continuous variables were evaluated using t-test or Mann-Whitney U test; categorical variables were evaluated using chi-square test or Fisher's exact test;

[0023] The feature analysis component performs multivariate logistic regression on the variables related to the outcome. When the P value of any variable in the multivariate logistic regression is less than 0.05, the variable is determined to be a significant variable affecting the outcome. The significant variable is included in the forward stepwise regression for screening to obtain the predictor of fullness.

[0024] The feature construction component determines the collinearity between the predictors of satiety through the variation inflation factor, excludes the variables with variation inflation factor greater than 2, and constructs the best feature set.

[0025] Further, the model verification module also includes a graphics drawing module and a model verification component;

[0026] The graphics drawing module draws a receiver operating characteristic curve, a calibration curve and a decision curve for the test set data to generate an internal validation curve;

[0027] The model verification component evaluates and verifies the internal verification curve generated by the graphics drawing module; specifically, by drawing the receiver operating characteristic curve, the area under the receiver operating characteristic curve is calculated to evaluate the distinguishing ability of the model; by calculating the calibration curve and the Hosmer-Lemeshow test, the calibration of the prediction model is evaluated; and by calculating the decision curve and the clinical impact curve, the clinical significance of the model is evaluated.

[0028] In a second aspect, the present invention further provides a method for predicting the risk of fullness in patients undergoing elective painless gastrointestinal endoscopy, the method comprising the following steps:

[0029] S1 is used to screen candidate predictors related to fullness based on the patient's clinical data, including the patient's demographic data, lifestyle variables, laboratory data, preoperative management data, and clinical characteristic data; the sample size is calculated based on the ratio of the number of individuals with outcome events to the number of candidate predictor variables; and the patient data information of the corresponding sample size is randomly retrieved as the original sample data;

[0030] S2, adding digital labels to the candidate predictive factors output by the sample data collection module and the satiety result indicators corresponding to the patient, generating a sample data set, and dividing the sample data set into a training set and a test set;

[0031] S3, feature selection was performed on the sample data in the training set. Specifically, the variables related to the satiety outcome indicators were screened out using univariate analysis, and the variables with significant univariate analysis were subjected to multivariate logistic regression. The variables with significant multivariate logistic regression were then included in the forward stepwise regression to screen out the predictors of satiety. Finally, the variation inflation factor was used to evaluate the collinearity between the predictors, and after excluding the predictors with significant collinearity, the optimal feature set was constructed.

[0032] S4, construct an individualized full stomach risk prediction function model before painless gastroenteroscopy: logit P = B1×P1+B2×P2+B3×P3+…Bn×Pn+constant, use the best feature set to train the individualized full stomach risk prediction function model, and obtain the full stomach prediction model;

[0033] S5, the test set is used to evaluate and validate the satiety prediction model.

[0034] The present invention adopts the above technical solution, and compared with the prior art, its significant technical effects are as follows:

[0035] 1. The stomach fullness risk prediction system for patients undergoing elective painless gastroenteroscopy of the present invention quantifies the probability values ​​of various predictive factors, objectively scores the gastric contents, and provides a visual and continuous scoring system to accurately reflect the risk probability in an individualized manner.

[0036] 2. The stomach fullness risk prediction system for patients undergoing elective painless gastrointestinal endoscopy of the present invention obtains the final variable that best explains the results and contributes the most to the evaluation by being easy to obtain, making the system more convenient and quick to use, by subjecting the best feature set to three processing steps of single variable screening, multivariate screening, and variation inflation factor screening to eliminate collinear variables.

[0037] 3. The stomach fullness risk prediction system for patients undergoing elective painless gastroenteroscopy of the present invention visualizes the contribution of various factors and the overall outcome through a nomogram, and the stomach fullness assessment results can be obtained through a mobile terminal, which greatly reduces the economic burden of users. It also uses characteristic curve evaluation indicators such as AUROC, calibration curve, DCA curve and CIC curve to perform multi-faceted test performance of sensitivity, specificity and accuracy, ensuring accurate evaluation of diverse sample data. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A flowchart of a system for predicting the risk of fullness in patients undergoing elective painless gastroenteroscopy was invented.

[0039] Figure 2 The present invention is a schematic representation of the baseline data of the system for predicting the risk of fullness in patients undergoing elective painless gastrointestinal endoscopy.

[0040] Figure 3 The present invention discloses a nomogram of the present full stomach of the system for predicting the risk of full stomach in patients undergoing elective painless gastrointestinal endoscopy.

[0041] Figure 4 This is a forest plot of the predictive factors of the stomach fullness risk prediction system for patients undergoing elective painless gastrointestinal endoscopy according to the present invention.

[0042] Figure 5 This is a schematic diagram of the summary of the results of the system for predicting the risk of fullness in patients undergoing elective painless gastrointestinal endoscopy according to the present invention. DETAILED DESCRIPTION

[0043] The present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0044] Figure 1 The invention is a flowchart of a system for predicting the risk of stomach fullness in patients undergoing elective painless gastrointestinal endoscopy. The embodiment of the invention discloses a system for predicting the risk of stomach fullness in patients undergoing elective painless gastrointestinal endoscopy. The stomach fullness risk assessment system includes a sample data collection module, a sample data processing module, a feature selection module, a model construction module and a model verification module.

[0045] The sample data collection module is used to screen candidate predictors related to fullness based on the patient's clinical data. The candidate predictors include the patient's demographic data, lifestyle variables, laboratory data, preoperative management data, and clinical characteristic data. The sample size is calculated based on the ratio of the number of individuals with outcome events to the number of candidate predictor variables.

[0046] The sample data processing module is used to add digital labels to the candidate prediction factors output by the sample data collection module and the satiety result indicators corresponding to the patient, generate a sample data set, and divide the sample data set into a training set and a test set.

[0047] The feature selection module is used to perform feature selection on the sample data in the training set. Specifically, the univariate analysis method is used to screen out the variables related to the satiety outcome indicators, and the variables with significant univariate analysis are subjected to multivariate logistic regression. The variables with significant multivariate logistic regression are then included in the forward stepwise regression to screen out the predictors of satiety. Finally, the variation inflation factor is used to evaluate the collinearity between the predictors, and after excluding the predictors with significant collinearity, the optimal feature set is constructed.

[0048] The model building module constructs an individualized full stomach risk prediction function model before painless gastroenteroscopy: logit P = B1×P1+B2×P2+B3×P3+…Bn×Pn+constant, and uses the best feature set to train the individualized full stomach risk prediction function model to obtain the full stomach prediction model.

[0049] The model validation module uses the test set to evaluate and validate the satiety prediction model.

[0050] 1. Sample data collection module

[0051] The sample data collection module is used to screen candidate predictors related to fullness based on the patient's clinical data. The candidate predictors include the patient's demographic data, lifestyle variables, laboratory data, preoperative management data, and clinical characteristic data. The sample size is calculated based on the ratio of the number of individuals with outcome events to the number of candidate predictor variables.

[0052] For example, in this embodiment, the sample data acquisition module is used to screen the patient's clinical data to obtain candidate predictive factors related to fullness, including age, BMI, gender, history of liver cirrhosis, history of cerebral infarction, history of diabetes, and history of gastroesophageal reflux. The data acquisition module randomly retrieves a certain amount of patient data information from the original data set of the data management center as sample data based on the patient data information input by the mobile terminal, and inputs it into the decision calculation module. Based on this embodiment, the ratio of the number of individuals with outcome events to the number of candidate predictive variables, that is, the number of events EPV of each variable is used to calculate the sample size. According to the empirical rule of multivariate logic, the EPV is taken as 10. It is known from previous evidence-based literature that the incidence of fullness in the painless gastroenteroscopy population, that is, the probability of positive events, is 12.2%. The number of predictive factors to be included is expected to be 8, and the calculation is performed as follows:

[0053]

[0054] The results show that at least 656 samples are needed to build a model. In this example, 1050 patients are randomly divided into a training set for analysis and construction of a satiety risk assessment model to meet the sample size requirement.

[0055] 2. Sample Data Processing Module

[0056] Figure 2 The present invention is a schematic diagram of the baseline data of the system for predicting the risk of gastric satiety in patients undergoing elective painless gastroenteroscopy. The sample data processing module is used to add digital labels to the candidate prediction factors output by the sample data acquisition module and the gastric satiety result indicators corresponding to the patient, generate a sample data set, and divide the sample data set into a training set and a test set.

[0057] For example, digital labels are added to candidate factors and outcome indicators to mark data attributes, such as 0 for male and 1 for female; 0 for fasting patients and 1 for full stomach patients. To avoid data leakage, the sample data set is first divided into training set and test set in a ratio of 7:3; variables with missing values ​​exceeding 25% in the training set are deleted, and variables with missing values ​​not exceeding 25% are interpolated to generate the following data: Figure 2 The baseline data table shown, in which continuous variables are imputed using the KNN method and categorical variables are imputed using the mode method.

[0058] 3. Feature Selection Module

[0059] The feature selection module is used to perform feature selection on the sample data in the training set. Specifically, the univariate analysis method is used to screen out the variables related to the satiety outcome indicators, and the variables with significant univariate analysis are subjected to multivariate logistic regression. The variables with significant multivariate logistic regression are then included in the forward stepwise regression to screen out the predictors of satiety. Finally, the variation inflation factor is used to evaluate the collinearity between the predictors, and after excluding the predictors with significant collinearity, the optimal feature set is constructed.

[0060] For example, in this implementation, univariate analysis is first used to screen out variables related to the outcome with P < 0.05, t-test or Mann-Whitney U test is used for continuous variables, and chi-square test or Fisher's exact test is used for categorical variables to evaluate differences; then, multivariate logistic regression is performed on variables that are significant in the univariate analysis; in order to be more applicable to this data set and the field of satiety prediction, variables with significant P < 0.05 in the multivariate logistic regression are further included in the forward stepwise regression to further screen out predictors of satiety; finally, the variation inflation factor VIF is used to evaluate the collinearity between variables. According to the number and characteristics of variables, this embodiment regards VIF < 2 as insignificant collinearity, and constructs the optimal feature set after excluding collinear variables.

[0061] 4. Model building module

[0062] This embodiment is constructed by R Studio software, and the optimal feature set is analyzed and predicted based on the logistic regression algorithm. The calculation formula is as follows:

[0063] logit P=B1×P1+B2×P2+B3×P3+…Bn×Pn+constant.

[0064] Among them, B is the satiety predictor and P is the variable.

[0065] The optimal feature set was used to train the individualized satiety risk prediction function model to obtain the satiety prediction model.

[0066] 5. Model Validation Module

[0067] Figure 3 The present invention discloses a nomogram of the present full stomach of the system for predicting the risk of full stomach in patients undergoing elective painless gastrointestinal endoscopy. Figure 4 This is a forest diagram of the predictive factors of the stomach fullness risk prediction system for patients undergoing elective painless gastroenteroscopy of the present invention. The model verification module uses a test set to evaluate and verify the stomach fullness prediction model. The present invention optimizes the traditional nomogram prediction model, which helps guide anesthesiologists in making anesthesia decisions and selecting fasting plans for patients undergoing painless gastroenteroscopy, optimizes the reasonable allocation of resources in the comfort medical management center, and significantly reduces the operating load of the endoscopy room.

[0068] In this embodiment, the full stomach prediction model is evaluated and verified by a test set through a model verification module. First, the receiver operating characteristic curve of the model is drawn, and the area under the receiver operating characteristic curve is calculated. The area under the receiver operating characteristic curve can evaluate the distinguishing ability of the model. The larger the area under the receiver operating characteristic curve, the better the distinguishing ability of the classifier. The calibration curve of the model is drawn, and the calibration of the prediction model can be evaluated based on the calibration curve and the Hosmer-Lemeshow test. The closer the calibration curve is to the 45° oblique line or Hosmer-Lemeshow P>0.05, the better the calibration ability of the model. The clinical significance of the model is evaluated by the decision curve and the clinical impact curve. The test set data is used to perform internal verification on the established model, and the accuracy of the prediction model is evaluated based on the internal verification curve. In addition, the optimal threshold of the nomogram model is determined on the ROC curve based on the Youden index, and the sensitivity, specificity and accuracy of each model are calculated under this threshold.

[0069] Figure 5 The present invention also provides a method for predicting the risk of fullness in patients undergoing elective painless gastrointestinal endoscopy, the method comprising the following steps:

[0070] S1, obtain the test patient data information, and randomly retrieve a certain amount of patient data information as the sample size through the sample data acquisition module.

[0071] S2. Add digital labels to the candidate predictive factors output by the sample data collection module and the satiety outcome indicators corresponding to the patient through the sample data processing module to generate a sample data set. Use the sample data processing module to divide the generated sample data set into a training set and a test set in a ratio of 7:3. Delete the variables with missing values ​​exceeding 25% in the training set, interpolate the variables with missing values ​​not exceeding 25%, and generate a baseline data table.

[0072] S3, using the feature selection module to perform feature selection on the sample data in the training set to construct the optimal feature set.

[0073] S4, inputting the best feature set obtained in step S3 into the trained satiety risk prediction function model to generate an evaluation result.

[0074] The above embodiments are only for illustrating the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A system for predicting the risk of stomach fullness in patients undergoing elective painless gastrointestinal endoscopy, characterized in that: The stomach satiety risk assessment system includes a sample data collection module, a sample data processing module, a feature selection module, a model construction module and a model verification module; The sample data collection module is used to screen candidate predictors related to fullness based on the patient's clinical data, the candidate predictors including the patient's demographic data, lifestyle variables, laboratory data, preoperative management data and clinical characteristic data; the sample size is calculated based on the ratio of the number of individuals with outcome events to the number of candidate predictor variables; The sample data processing module is used to add digital labels to the candidate prediction factors output by the sample data acquisition module and the satiety result indicators corresponding to the patient, generate a sample data set, and divide the sample data set into a training set and a test set; The feature selection module is used to perform feature selection on the sample data in the training set. Specifically, the variables related to the satiety outcome index are screened out using the univariate analysis method, and the variables with significant univariate analysis are subjected to multivariate logistic regression. The variables with significant multivariate logistic regression are then included in the forward stepwise regression to screen out the predictors of satiety. Finally, the variation inflation factor is used to evaluate the collinearity between the predictors, and after excluding the predictors with significant collinearity, the optimal feature set is constructed. The model building module builds an individualized full stomach risk prediction function model before painless gastroenteroscopy: logit P = B1×P1+B2×P2+B3×P3+…Bn×Pn+constant, and uses the best feature set to train the individualized full stomach risk prediction function model to obtain a full stomach prediction model; The model validation module uses a test set to evaluate and validate the satiety prediction model.

2. The system for predicting the risk of fullness in patients undergoing elective painless gastrointestinal endoscopy according to claim 1, characterized in that: The demographic characteristics include patient age, body mass index and gender; lifestyle variables include patient smoking history and drinking history; laboratory data include hemoglobin; preoperative management data include preoperative fasting time and preoperative diet type; clinical characteristics include history of mental illness, hypothyroidism, diabetes, cirrhosis, chronic renal dysfunction, cerebral infarction, sequelae of cerebral infarction, history of PPI or H2 receptor antagonist use, history of prokinetic drugs use and type of endoscopy to be examined.

3. The system for predicting the risk of fullness in patients undergoing elective painless gastrointestinal endoscopy according to claim 1, characterized in that: The sample data collection module includes a sample size calculation component; The sample size calculation component calculates the sample size for each variable based on the empirical rule of multivariate logic, and the calculation formula is as follows:

4. The system for predicting the risk of fullness in patients undergoing elective painless gastrointestinal endoscopy according to claim 1, characterized in that: The sample data processing module includes a sample division component, a sample deletion component and a sample interpolation component; The sample partitioning component partitions the sample data set into a training set and a test set in a ratio of 7:3; The sample deletion component is used to delete variables with more than 25% missing values ​​in the training set; The sample interpolation component interpolates variables with missing values ​​not exceeding 25% to generate a baseline data table; wherein continuous variables are interpolated using the KNN method, and categorical variables are interpolated using the majority method.

5. The system for predicting the risk of fullness in patients undergoing elective painless gastrointestinal endoscopy according to claim 1, characterized in that: The feature selection module includes a feature screening component, a feature analysis component and a feature construction component; The feature screening component was screened using univariate analysis to screen out variables with a P value of <0.05 in the training set, which were defined as variables associated with the outcome; among them, continuous variables were evaluated using t-test or Mann-Whitney U test; categorical variables were evaluated using chi-square test or Fisher's exact test; The feature analysis component performs multivariate logistic regression on the variables related to the outcome. When the P value of any variable in the multivariate logistic regression is less than 0.05, the variable is determined to be a significant variable affecting the outcome. The significant variable is included in the forward stepwise regression for screening to obtain the predictor of fullness. The feature construction component determines the collinearity between satiety prediction factors through the variation inflation factor, excludes variables with variation inflation factors greater than 2, and constructs the best feature set.

6. The system for predicting the risk of fullness in patients undergoing elective painless gastrointestinal endoscopy according to claim 1, characterized in that: The model verification module also includes a graphics drawing module and a model verification component; The graphics drawing module draws a receiver operating characteristic curve, a calibration curve and a decision curve for the test set data to generate an internal validation curve; The model verification component evaluates and verifies the internal verification curve generated by the graphics drawing module; specifically, by drawing the receiver operating characteristic curve, the area under the receiver operating characteristic curve is calculated to evaluate the distinguishing ability of the model; by calculating the calibration curve and the Hosmer-Lemeshow test, the calibration of the prediction model is evaluated; and by calculating the decision curve and the clinical impact curve, the clinical significance of the model is evaluated.

7. A method for predicting the risk of fullness in patients undergoing elective painless gastrointestinal endoscopy, characterized in that: The prediction method comprises the following steps: S1 is used to screen candidate predictors related to fullness based on the patient's clinical data, including the patient's demographic data, lifestyle variables, laboratory data, preoperative management data, and clinical characteristic data; the sample size is calculated based on the ratio of the number of individuals with outcome events to the number of candidate predictor variables; and the patient data information of the corresponding sample size is randomly retrieved as the original sample data; S2, adding digital labels to the candidate predictive factors output by the sample data collection module and the satiety result indicators corresponding to the patient, generating a sample data set, and dividing the sample data set into a training set and a test set; S3, feature selection was performed on the sample data in the training set. Specifically, the variables related to the satiety outcome indicators were screened out using univariate analysis, and the variables with significant univariate analysis were subjected to multivariate logistic regression. The variables with significant multivariate logistic regression were then included in the forward stepwise regression to screen out the predictors of satiety. Finally, the variation inflation factor was used to evaluate the collinearity between the predictors, and after excluding the predictors with significant collinearity, the optimal feature set was constructed. S4, construct an individualized full stomach risk prediction function model before painless gastroenteroscopy: logit P = B1×P1+B2×P2+B3×P3+…Bn×Pn+constant, use the best feature set to train the individualized full stomach risk prediction function model, and obtain the full stomach prediction model; S5, the test set is used to evaluate and validate the satiety prediction model.

Images