Method for predicting acute postoperative pain risk of gastric cancer patient based on biomarker

Abstract

The invention relates to the technical field of medical data management, and particularly discloses a gastric cancer patient acute postoperative pain risk prediction method based on a biomarker. The method is used for solving the problem that a method for predicting the acute postoperative pain risk of a patient according to cancer tissue methylation information does not exist in existing medical management equipment for acute postoperative pain of a gastric cancer patient. Comprising the steps of obtaining primary screening biomarker information, obtaining secondary screening biomarker information, training an acute postoperative pain risk prediction model, obtaining target biomarker information and obtaining a target pain event outcome of a target patient. According to the method, through stage feature screening, fusion of biomarker feature cancer tissue methylation information, accurate screening of similar patients and machine learning model optimization, accurate prediction of postoperative pain risks is realized, the sensitivity, specificity and clinical applicability of the model are improved, and scientific basis and technical support are provided for personalized pain management.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data management, and more specifically, to a method for predicting the risk of acute postoperative pain in gastric cancer patients based on biomarkers. Background Art

[0002] Acute postoperative pain refers to pain caused by acute trauma, incision and visceral organ damage, irritation and stimulation of drainage caused by the surgical process. It often occurs 24 to 48 hours after surgery and can have a very serious adverse effect on the prognosis of surgical patients. These effects include immunosuppression caused by unrelieved pain (delayed wound healing, prolonged recovery time, and increased risk of postoperative infection), psychological effects including anxiety and depression, delayed time to get out of bed, increased risk of venous thromboembolism and pulmonary infection, etc. Patients with inadequate postoperative pain management have an increased risk of postoperative chronic pain (still having pain three months after surgery) and the need for readmission. For patients after gastrointestinal tumor surgery, the management of acute postoperative pain can reduce immunosuppression and promote early bed rest, which is of great significance for the recovery of gastrointestinal function, especially for gastric cancer patients. Ding Lirong et al. from the First Affiliated Huainan Hospital of Nanjing Medical University mentioned in the public document "Correlation between Acute Postoperative Pain and Pain Sensitivity in Patients Undergoing Laparoscopic Gastrointestinal Tumor Surgery" that they conducted a pain sensitivity analysis on patients undergoing laparoscopic gastrointestinal cancer surgery under general anesthesia in the elective period, and pointed out that pain sensitivity Sensitivity is related to the occurrence of acute postoperative pain in patients undergoing laparoscopic gastrointestinal cancer surgery, which seriously affects the prognosis of patients. In Pan Ranran's thesis "Screening of DNA methylation diagnostic biomarkers for multiple candidate genes in gastrointestinal cancer and their functional research", the methylation levels of 26 candidate genes were detected in gastrointestinal cancer tissues by fluorescent quantitative methylation-specific PCR, and it was pointed out that the promoter methylation values ​​of ALOX12, EPAS1, FANCF, HIF1A, IL10, KDM1A, KLF5, LEPR, MTAP, PTPN11, PRMT6, RBL2, and TNFAIP3 in cancer tissues of gastric cancer patients were significantly lower than those in adjacent cancer tissues, while the methylation values ​​of CXCL12, EBF3, GPX3, HHIP, HOXA9, and TLR2 promoters in cancer tissues were significantly higher than those in adjacent cancer tissues. However, the current medical management equipment for acute postoperative pain in gastric cancer patients does not yet have a method to predict the risk of acute postoperative pain in patients based on the expression information of LRRTM, and cannot provide effective data support for the existing medical management equipment for acute postoperative pain in gastric cancer patients. Summary of the invention

[0003] In order to overcome the above-mentioned defects of the prior art, the present invention provides a biomarker-based method for predicting the risk of acute postoperative pain in gastric cancer patients. Through stage feature screening, fusion of biomarker features, precise screening of similar patients and machine learning model optimization, accurate prediction of postoperative pain risk is achieved, the sensitivity, specificity and clinical applicability of the model are improved, and a scientific basis and technical support are provided for personalized pain management.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] The biomarker-based method for predicting the risk of acute postoperative pain in patients with gastric cancer includes the following steps:

[0006] Step 1, obtaining the initial screening biomarker information: obtaining the historical patient's initial biomarker characteristic values ​​and pain event outcomes from the hospital's historical pain management records of all gastric cancer patients, obtaining the initial screening biomarker information, and performing feature screening on the initial biomarker characteristics to obtain the target biomarker characteristics, wherein the initial biomarker characteristics include basic characteristics and pain management characteristics, the basic characteristics include clinical monitoring data characteristics, past medical history data characteristics, and drug use history data characteristics, the pain management characteristics include the patient's pain assessment index and pain management plan at the time of admission, and the pain event outcomes include whether the patient has experienced an acute postoperative pain event or not within the observation time window;

[0007] Step 2, obtaining the second screening biomarker information: screening the target biomarker characteristic values ​​and pain event outcomes of each historical patient from the primary screening biomarker information to obtain the second screening biomarker information;

[0008] Step 3, training an acute postoperative pain risk prediction model: using the secondary screening biomarker information to train an acute postoperative pain risk prediction model to obtain an initial target prediction model;

[0009] Step 4, obtaining target biomarker information: obtaining biomarker information of the target patient, and obtaining target biomarker information according to the target biomarker characteristics;

[0010] Step 5, obtaining the target pain event outcome of the target patient: input the target biomarker information into the initial target prediction model to obtain the target pain event outcome of the target patient.

[0011] As a further embodiment of the present invention, in step 1, the process of screening the initial biomarker features includes:

[0012] Step 11, splitting the unit stage and the feature screening sub-stage: splitting the screening process of the initial screening biomarker information into the unit stage, and the unit stage includes the feature screening sub-stage;

[0013] Step 12, obtaining the candidate screening process: obtaining the preset screening method of each unit stage, the number of preset screening methods corresponding to the feature screening sub-stage is at least 2 (different screening methods perform feature selection based on different assumptions and algorithms. If only one method is used, it is easy to be affected by the limitations of the method, resulting in the selected features not being universal and unable to capture complex patterns in the data). The preset screening methods of all unit stages are combined to obtain the candidate screening process;

[0014] Step 13, obtaining the target biomarker characteristics of the target patient: processing the initial screening biomarker information according to each candidate screening process to obtain the corresponding target biomarker characteristics.

[0015] As a further embodiment of the present invention, in step 3, the process of using the secondary screening biomarker information to train an acute postoperative pain risk prediction model includes:

[0016] Step 31, establishing an acute postoperative pain risk prediction model: establishing an acute postoperative pain risk prediction model based on the target biomarker characteristics corresponding to the second screening biomarker information and a preset machine learning model;

[0017] Step 32, obtaining an initial candidate prediction model: training each acute postoperative pain risk prediction model according to the second screening biomarker information to obtain an initial candidate prediction model;

[0018] Step 33, obtaining an initial target prediction model: evaluating the performance indicators of the initial candidate prediction models, and obtaining the initial target prediction model according to the initial candidate prediction model that best corresponds to the performance indicator.

[0019] As a further solution of the present invention, in step 33, the process of obtaining the initial target prediction model according to the initial candidate prediction model that best corresponds to the performance indicator includes:

[0020] Step 331, obtaining a second optimal target prediction model: taking the initial candidate prediction model with the best performance index as the second optimal target prediction model;

[0021] Step 332, obtaining target biomarker features according to the input features of the second best target prediction model: determining whether the number of input features of the second best target prediction model is greater than a first preset threshold, if the number of input features of the second best target prediction model is greater than the first preset threshold, arranging the input features of the second best target prediction model in descending order according to the importance of the biomarker features, and taking the biomarker features of all input features of the second best target prediction model whose ranking is less than or equal to the first preset threshold as the target biomarker features;

[0022] Step 333, obtaining a third screening biomarker feature: screening the second screening biomarker information input into the second optimal target prediction model according to the target biomarker feature to obtain third screening biomarker information;

[0023] Step 334, obtaining a second candidate prediction model: establishing an acute postoperative pain risk prediction model based on the three-screening biomarker information and a preset machine learning model, and using the three-screening biomarker information to train each acute postoperative pain risk prediction model to obtain a second candidate prediction model;

[0024] Step 335, obtaining an initial target prediction model: evaluating the performance indicators of all second candidate prediction models, and taking the second candidate prediction model with the best performance indicator as the initial target prediction model.

[0025] As a further embodiment of the present invention, in step 332, the importance of the biomarker feature includes the degree of relevance to the outcome of the pain event and the degree of predictive contribution;

[0026] As a further solution of the present invention, in step 322, the process of arranging the inputs of the second-best target prediction model in descending order according to the importance of the biomarker features includes:

[0027] Step 3321, obtaining the correlation degree of the pain event outcome: obtaining the correlation degree between each input feature of the two-optimal target prediction model and the pain event outcome to obtain the correlation degree of the pain event outcome, wherein the correlation degree between the input feature and the pain event outcome is obtained according to the preset screening method corresponding to the two-optimal target prediction model in the feature screening sub-stage;

[0028] Step 3322, obtaining the prediction contribution degree: respectively calculating the contribution degree of each input feature of the two-optimal target prediction model to the prediction operation completed by the two-optimal target prediction model, and obtaining the prediction contribution degree;

[0029] Step 3323, obtaining the importance of the input feature: the importance of the input feature is calculated according to the relevance degree of the pain event outcome and the prediction contribution degree obtained in steps 3321 and 3322 respectively.

[0030] As a further solution of the present invention, in step 334, the construction process of the second candidate prediction model includes:

[0031] Step 3341, obtaining four-screening biomarker information: obtaining the value of the third biomarker characteristic of the gastric cancer patient in the historical medical records, and obtaining the four-screening biomarker information;

[0032] Step 3342, obtaining input features of the second candidate prediction model: using the third biomarker feature and the biomarker features corresponding to the three screening biomarker information as input features of the second candidate prediction model;

[0033] Step 3343, obtaining a second candidate prediction model: using the four-screening biomarker information and the three-screening biomarker information to train an acute postoperative pain risk prediction model to obtain a second candidate prediction model.

[0034] As a further embodiment of the present invention, the third biomarker feature includes cancer tissue gene methylation information, which is calculated based on the promoter methylation values ​​of ALOX12, EPAS1, FANCF, HIF1A, IL10, KDM1A, KLF5, LEPR, MTAP, PTPN11, PRMT6, RBL2, and TNFAIP3 of gastric cancer patients in the pain management history records of gastric cancer patients in the hospital and the cancer tissue methylation values ​​of CXCL12, EBF3, GPX3, HHIP, HOXA9, and TLR2 promoters;

[0035] Promoter methylation values ​​of ALOX12, EPAS1, FANCF, HIF1A, IL10, KDM1A, KLF5, LEPR, MTAP, PTPN11, PRMT6, RBL2, and TNFAIP3 in cancer tissues obtained by including but not limited to tissue examination at 24 hours and 48 hours after surgery in historical gastric cancer patients, obtained by gene expression profiling;

[0036] The cancer tissue methylation values ​​of CXCL12, EBF3, GPX3, HHIP, HOXA9, and TLR2 promoters include the cancer tissue methylation values ​​of CXCL12, EBF3, GPX3, HHIP, HOXA9, and TLR2 promoters in cancer tissues of historical gastric cancer patients obtained by including but not limited to tissue examination 24 hours and 48 hours after surgery.

[0037] As a further scheme of the present invention, the process of obtaining the gene methylation information of cancer tissues of historical patients is calculated based on the promoter methylation values ​​of ALOX12, EPAS1, FANCF, HIF1A, IL10, KDM1A, KLF5, LEPR, MTAP, PTPN11, PRMT6, RBL2, and TNFAIP3 of gastric cancer patients in the pain management history records of gastric cancer patients in the hospital and the cancer tissue methylation values ​​of CXCL12, EBF3, GPX3, HHIP, HOXA9, and TLR2 promoters, including:

[0038] Step 41, obtaining actual cancer tissue gene methylation information: according to the promoter methylation values ​​of ALOX12, EPAS1, FANCF, HIF1A, IL10, KDM1A, KLF5, LEPR, MTAP, PTPN11, PRMT6, RBL2, and TNFAIP3 of historical gastric cancer patients and the cancer tissue methylation values ​​of the promoters of CXCL12, EBF3, GPX3, HHIP, HOXA9, and TLR2, the cancer tissue gene methylation information of the corresponding historical patients is calculated;

[0039] Step 42, obtaining reference cancer tissue gene methylation information: querying a database according to the initial screening biomarker information of historical patients to obtain reference patients, the reference patients being historical patients whose initial screening biomarker information similarity is greater than a second preset threshold among all historical patients who did not experience acute postoperative pain events within the observation time window, calculating the mean of the actual cancer tissue gene methylation information of all reference patients, and using the mean of the actual cancer tissue gene methylation information of all reference patients as the reference cancer tissue gene methylation information of the historical patients;

[0040] Step 43, obtaining the cancer tissue gene methylation information of the historical patient: calculating the ratio of the actual cancer tissue gene methylation information to the reference cancer tissue gene methylation information, and using the ratio as the cancer tissue gene methylation information of the historical patient.

[0041] As a further solution of the present invention, in step 42, the process of querying the database to obtain reference patients based on the initial screening biomarker information of historical patients includes:

[0042] Step 421, obtaining basic similarity: calculating the similarity between the value of the pain assessment index of each historical patient and the value of the basic feature of other historical patients, and obtaining the basic similarity;

[0043] Step 422, obtaining the similarity of pain management plans: calculating the similarity between the pain management plan of each historical patient and the pain management plans of other historical patients respectively, to obtain the similarity of pain management plans;

[0044] Step 423, obtaining the similarity of the pain assessment index: calculating the similarity between the value of the pain assessment index of each historical patient and the value of the assessment index of other historical patients, and obtaining the similarity of the pain assessment index;

[0045] Step 424, obtaining fusion similarity: linearly weighting the basic similarity, pain management indicator similarity and pain management plan similarity, and obtaining the fusion similarity between each historical patient and other historical patients based on the values ​​corrected by the historical patient's physiological monitoring parameters, wherein the weight of the basic similarity and the weight of the pain management plan similarity are greater than the weight of the indicator similarity, and the pain assessment indicator is obtained through the patient's report results. The patient records the pain data through the smart device and shares it between the patient and the hospital.

[0046] In order to solve the technical problems raised by the background technology, the technical effect of the method proposed in the present invention is as follows: the present invention can predict the risk of acute postoperative pain in gastric cancer patients by combining the patient's clinical characteristics, pain management records and biomarker data, identify key biomarker features that are closely related to postoperative pain through multi-stage feature screening and data processing, and then provide personalized pain risk assessment by establishing a machine learning model. By obtaining and screening the initial screening biomarker information, redundant information is minimized to the greatest extent, and factors that have an impact on pain prediction are retained, thereby ensuring the efficiency and accuracy of the model. The feature selection strategy is further optimized using the candidate screening process, and the robustness and generalization ability of the model are improved through diversified screening methods. It can not only provide data support for postoperative pain management and help formulate personalized pain management plans, but also reduce the inhibitory effect of pain on the immune system and the risk of postoperative infection through early prediction of pain risks, promote early recovery of patients and recovery of gastrointestinal function, significantly improve the quality of life of patients and reduce the rate of readmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flow chart of the method proposed by the present invention;

[0048] Figure 2 A flow chart for feature screening of initial biomarker features for the present invention;

[0049] Figure 3 A flowchart of the present invention for training an acute postoperative pain risk prediction model using two-screen biomarker information;

[0050] Figure 4 A flowchart of the present invention for obtaining an initial target prediction model based on an initial candidate prediction model with the best performance indicator;

[0051] Figure 5 A flowchart of the present invention for arranging the inputs of the second-best target prediction model in descending order according to the importance of the biomarker features;

[0052] Figure 6 is a flow chart for constructing the second candidate model of the present invention;

[0053] Figure 7 This is a flow chart of the present invention for obtaining historical patient LRRTM1 expression information based on the LRRTM1 gene expression level and LRRTM1 protein expression level of the patient in the hospital's esophageal cancer patient pain management history record;

[0054] Figure 8 The present invention is a flow chart for querying a database to obtain reference patients based on the initial screening biomarker information of historical patients. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] Example 1

[0057] like Figure 1 The method for predicting the risk of acute postoperative pain in gastric cancer patients based on biomarkers proposed by the present invention comprises the following steps:

[0058] Step 1, obtaining the initial screening biomarker information: obtaining the historical patient's initial biomarker characteristic values ​​and pain event outcomes from the hospital's historical pain management records of all gastric cancer patients, obtaining the initial screening biomarker information, and performing feature screening on the initial biomarker characteristics to obtain the target biomarker characteristics, wherein the initial biomarker characteristics include basic characteristics and pain management characteristics, the basic characteristics include clinical monitoring data characteristics, past medical history data characteristics, and drug use history data characteristics, the pain management characteristics include the patient's pain assessment index and pain management plan at the time of admission, and the pain event outcomes include whether the patient has experienced an acute postoperative pain event or not within the observation time window;

[0059] Step 2, obtaining the second screening biomarker information: screening the target biomarker characteristic values ​​and pain event outcomes of each historical patient from the primary screening biomarker information to obtain the second screening biomarker information;

[0060] Step 3, training an acute postoperative pain risk prediction model: using the secondary screening biomarker information to train an acute postoperative pain risk prediction model to obtain an initial target prediction model;

[0061] Step 4, obtaining target biomarker information: obtaining biomarker information of the target patient, and obtaining target biomarker information according to the target biomarker characteristics;

[0062] Step 5, obtaining the target pain event outcome of the target patient: input the target biomarker information into the initial target prediction model to obtain the target pain event outcome of the target patient.

[0063] The present invention can predict the risk of acute postoperative pain in gastric cancer patients by combining the patient's clinical characteristics, pain management records and biomarker data. Through multi-stage feature screening and data processing, key biomarker features that are closely related to postoperative pain are identified, and then a machine learning model is established to provide personalized pain risk assessment. By obtaining and screening features from initial screening biomarker information, redundant information is minimized, and factors that have an impact on pain prediction are retained, ensuring the efficiency and accuracy of the model. The feature selection strategy is further optimized using the candidate screening process, and the robustness and generalization ability of the model are improved through diversified screening methods. It can not only provide data support for postoperative pain management and help formulate personalized pain management plans, but also reduce the inhibitory effect of pain on the immune system and the risk of postoperative infection through early prediction of pain risks, promote early recovery of patients and recovery of gastrointestinal function, significantly improve the quality of life of patients and reduce the rate of readmission.

[0064] Example 2

[0065] like Figure 2 As shown, it is the process of characteristic screening of initial biomarkers in step 1 of the method proposed in the present invention.

[0066] In step 1, the process of feature screening for the initial biomarker signature includes:

[0067] Step 11, splitting the unit stage and the feature screening sub-stage: splitting the screening process of the initial screening biomarker information into the unit stage, and the unit stage includes the feature screening sub-stage;

[0068] Step 12, obtaining a candidate screening process: obtaining a preset screening method for each unit stage, the number of preset screening methods corresponding to the feature screening sub-stage is at least 2, and the preset screening methods of all unit stages are combined to obtain a candidate screening process;

[0069] Step 13, obtaining the target biomarker characteristics of the target patient: processing the initial screening biomarker information according to each candidate screening process to obtain the corresponding target biomarker characteristics.

[0070] This embodiment constructs multiple candidate screening processes by splitting the initial biomarker feature screening process into multiple unit stages and setting at least two preset screening methods for each feature screening sub-stage. It improves the diversity and robustness of feature selection by combining different screening paths, effectively avoids the deviation that may be caused by a single screening method, enhances the adaptability of the model to different data features, and ensures that the selected target biomarker features have strong predictive ability and generalization performance, ultimately achieving more accurate prediction of acute postoperative pain risk, improving the accuracy of feature screening, reducing the risk of overfitting, and providing reliable data support for subsequent clinical applications.

[0071] Example 3

[0072] like Figure 3 As shown, the process of using the new information of the second screening biomarker to train the postoperative pain risk prediction model in step 3 of the present invention.

[0073] In step 3, the process of training the acute postoperative pain risk prediction model using the secondary screening biomarker information includes:

[0074] Step 31, establishing an acute postoperative pain risk prediction model: establishing an acute postoperative pain risk prediction model based on the target biomarker characteristics corresponding to the second screening biomarker information and a preset machine learning model;

[0075] Step 32, obtaining an initial candidate prediction model: training each acute postoperative pain risk prediction model according to the second screening biomarker information to obtain an initial candidate prediction model;

[0076] Step 33, obtaining an initial target prediction model: evaluating the performance indicators of the initial candidate prediction models, and obtaining the initial target prediction model according to the initial candidate prediction model that best corresponds to the performance indicator.

[0077] By using the secondary screening biomarker information to train the acute postoperative pain risk prediction model in step 3, and by evaluating the performance indicators of multiple candidate models, the best model is finally selected as the initial target prediction model, which can improve the accuracy and stability of the prediction model. Through multiple rounds of screening and performance evaluation, it is ensured that the selected prediction model has the best prediction ability, thereby effectively coping with the individual differences in acute postoperative pain, providing a more accurate postoperative pain risk prediction, and providing a scientific basis for the clinical formulation of personalized pain management plans, effectively improving patients' postoperative recovery and prognosis.

[0078] It should be noted that if Figure 4 As shown, in step 33, the process of obtaining the initial target prediction model according to the initial candidate prediction model that best corresponds to the performance indicator includes:

[0079] Step 331, obtaining a second optimal target prediction model: taking the initial candidate prediction model with the best performance index as the second optimal target prediction model;

[0080] Step 332, obtaining target biomarker features according to the input features of the second best target prediction model: determining whether the number of input features of the second best target prediction model is greater than a first preset threshold, if the number of input features of the second best target prediction model is greater than the first preset threshold, arranging the input features of the second best target prediction model in descending order according to the importance of the biomarker features, and taking the biomarker features of all input features of the second best target prediction model whose ranking is less than or equal to the first preset threshold as the target biomarker features;

[0081] Step 333, obtaining a third screening biomarker feature: screening the second screening biomarker information input into the second optimal target prediction model according to the target biomarker feature to obtain third screening biomarker information;

[0082] Step 334, obtaining a second candidate prediction model: establishing an acute postoperative pain risk prediction model based on the three-screening biomarker information and a preset machine learning model, and using the three-screening biomarker information to train each acute postoperative pain risk prediction model to obtain a second candidate prediction model;

[0083] Step 335, obtaining an initial target prediction model: evaluating the performance indicators of all second candidate prediction models, and taking the second candidate prediction model with the best performance indicator as the initial target prediction model.

[0084] In step 33, the initial candidate prediction model with the best performance index is used as the basis to further optimize the feature selection and model training process, and the second-best target prediction model is used to screen out the target biomarker features with high correlation. The second candidate prediction model is retrained and evaluated based on the three-screened biomarker information, and finally the model with the best performance is selected as the initial target prediction model, thereby significantly improving the prediction accuracy, robustness and clinical applicability of the model, solving the problems of redundant feature interference and insufficient model generalization ability, and providing reliable data support for accurate postoperative pain risk prediction and personalized pain management.

[0085] It should be noted that in step 332, the importance of the biomarker feature includes the degree of relevance to the outcome of the pain event and the degree of predictive contribution.

[0086] like Figure 5 As shown, in step 322, the process of arranging the inputs of the second-best target prediction model in descending order according to the importance of the biomarker features includes:

[0087] Step 3321, obtaining the correlation degree of the pain event outcome: obtaining the correlation degree between each input feature of the two-optimal target prediction model and the pain event outcome to obtain the correlation degree of the pain event outcome, wherein the correlation degree between the input feature and the pain event outcome is obtained according to the preset screening method corresponding to the two-optimal target prediction model in the feature screening sub-stage;

[0088] Step 3322, obtaining the prediction contribution degree: respectively calculating the contribution degree of each input feature of the two-optimal target prediction model to the prediction operation completed by the two-optimal target prediction model, and obtaining the prediction contribution degree;

[0089] Step 3323, obtaining the importance of the input feature: the importance of the input feature is calculated according to the relevance degree of the pain event outcome and the prediction contribution degree obtained in steps 3321 and 3322 respectively.

[0090] It should be noted that, in step 3321, the correlation degree of the outcome of the pain event is obtained, and the formula is:

[0091] R i =Corr(X i ,Y)

[0092] Where: i is the input feature index, X i is the i-th input feature of the two-optimal target prediction model, Y is the pain event outcome variable, which is a binary variable (1 indicates the occurrence of acute postoperative pain event, 0 indicates no acute postoperative pain event), Corr is the correlation calculation function, R i For feature X i The degree of correlation with the pain event outcome Y.

[0093] It should be noted that, in step 3322, the formula for predicting the contribution degree is:

[0094]

[0095] Where: C i For feature X i The prediction contribution of n is the total number of input features. For feature X i The contribution of the second-best target prediction name to the overall prediction accuracy is expressed using the SHAP value (Shapley Additive Explanations) of feature X i The marginal contribution to the model's predictions.

[0096] It should be noted that the importance formula of the input feature is:

[0097] I i =αR i +βCi

[0098] Where: I i For feature X i The importance of is, α, β are weight parameters, which are set according to expert experience.

[0099] By calculating the importance of biomarker features through the above steps, combining the relevance of pain event outcomes and the degree of predictive contribution, and comprehensively calculating the importance of input features through weight parameters, the most critical biomarker features for acute postoperative pain risk prediction can be accurately identified, effectively combining the advantages of statistics and machine learning models, considering both the direct correlation between features and target variables, and comprehensively evaluating the actual contribution of features in the prediction model, thereby ensuring that the selected features have high biological significance and predictive performance. In addition, the marginal contribution of features is accurately quantified using SHAP values, further improving the interpretability and reliability of the model, providing a solid foundation for building a more accurate postoperative pain risk prediction model, and ultimately optimizing the feature screening process and improving the prediction accuracy and robustness of the model.

[0100] like Figure 6 As shown, in step 334, the construction process of the second candidate prediction model includes:

[0101] Step 3341, obtaining four-screening biomarker information: obtaining the value of the third biomarker characteristic of the gastric cancer patient in the historical medical records, and obtaining the four-screening biomarker information;

[0102] Step 3342, obtaining input features of the second candidate prediction model: using the third biomarker feature and the biomarker features corresponding to the three screening biomarker information as input features of the second candidate prediction model;

[0103] Step 3343, obtaining a second candidate prediction model: using the four-screening biomarker information and the three-screening biomarker information to train an acute postoperative pain risk prediction model to obtain a second candidate prediction model.

[0104] In step 334, by constructing a second candidate prediction model, the third biomarker feature is combined with the three-screening biomarker feature, which further enriches the diversity and accuracy of the input features and improves the prediction ability of the model. The introduction of the four-screening biomarker information is based on higher-dimensional data, ensuring that the model captures more detailed biological associations. By integrating multiple rounds of screening features, the input variables of the model are more biologically meaningful and clinically relevant, effectively reducing the interference of noise data. At the same time, multiple rounds of training are used to optimize the model parameters, which significantly improves the prediction accuracy and robustness of the model. It not only solves the possible omission problem in the initial model feature selection, but also enhances the model's ability to accurately predict the patient's risk of acute postoperative pain, providing more reliable technical support for personalized pain management.

[0105] Example 4

[0106] The third biomarker feature includes the gene methylation information of cancer tissues, which is calculated based on the promoter methylation values ​​of ALOX12, EPAS1, FANCF, HIF1A, IL10, KDM1A, KLF5, LEPR, MTAP, PTPN11, PRMT6, RBL2, and TNFAIP3 of gastric cancer patients in the pain management history records of gastric cancer patients in the hospital, and the cancer tissue methylation values ​​of CXCL12, EBF3, GPX3, HHIP, HOXA9, and TLR2 promoters;

[0107] Promoter methylation values ​​of ALOX12, EPAS1, FANCF, HIF1A, IL10, KDM1A, KLF5, LEPR, MTAP, PTPN11, PRMT6, RBL2, and TNFAIP3 in cancer tissues obtained by including but not limited to tissue examination at 24 hours and 48 hours after surgery in historical gastric cancer patients, obtained by gene expression profiling;

[0108] The cancer tissue methylation values ​​of CXCL12, EBF3, GPX3, HHIP, HOXA9, and TLR2 promoters include the cancer tissue methylation values ​​of CXCL12, EBF3, GPX3, HHIP, HOXA9, and TLR2 promoters in cancer tissues of historical gastric cancer patients obtained by including but not limited to tissue examination 24 hours and 48 hours after surgery.

[0109] By introducing the third biomarker feature and using gene expression spectrum analysis, we can accurately obtain the methylation value expression levels of cancer tissues in historical gastric cancer patients 24 hours and 48 hours after surgery, which can reveal the potential mechanisms related to the risk of acute postoperative pain from the molecular biology level, provide a more comprehensive biological basis, and significantly improve the sensitivity and specificity of the prediction model for the risk of acute postoperative pain. By introducing the methylation value expression level characteristics of cancer tissues, we can not only make up for the shortcomings of traditional clinical characteristics and general biomarkers, but also enhance the biological interpretability of the model, provide technical support for the precise screening of key features that are highly correlated with pain risk, effectively solve the problem of insufficient utilization of molecular level information by existing pain management equipment, and provide strong support for optimizing postoperative pain risk prediction and personalized management.

[0110] like Figure 7 As shown, according to the promoter methylation values ​​of ALOX12, EPAS1, FANCF, HIF1A, IL10, KDM1A, KLF5, LEPR, MTAP, PTPN11, PRMT6, RBL2, and TNFAIP3 of gastric cancer patients in the hospital's gastric cancer patient pain management history records and the cancer tissue methylation values ​​of CXCL12, EBF3, GPX3, HHIP, HOXA9, and TLR2 promoters, the process of obtaining the gene methylation information of the cancer tissue of the historical patients includes:

[0111] Step 41, obtaining actual cancer tissue gene methylation information: according to the promoter methylation values ​​of ALOX12, EPAS1, FANCF, HIF1A, IL10, KDM1A, KLF5, LEPR, MTAP, PTPN11, PRMT6, RBL2, and TNFAIP3 of historical gastric cancer patients and the cancer tissue methylation values ​​of the promoters of CXCL12, EBF3, GPX3, HHIP, HOXA9, and TLR2, the cancer tissue gene methylation information of the corresponding historical patients is calculated;

[0112] Step 42, obtaining reference cancer tissue gene methylation information: querying a database according to the initial screening biomarker information of historical patients to obtain reference patients, the reference patients being historical patients whose initial screening biomarker information similarity is greater than a second preset threshold among all historical patients who did not experience acute postoperative pain events within the observation time window, calculating the mean of the actual cancer tissue gene methylation information of all reference patients, and using the mean of the actual cancer tissue gene methylation information of all reference patients as the reference cancer tissue gene methylation information of the historical patients;

[0113] Step 43, obtaining the cancer tissue gene methylation information of the historical patient: calculating the ratio of the actual cancer tissue gene methylation information to the reference cancer tissue gene methylation information, and using the ratio as the cancer tissue gene methylation information of the historical patient.

[0114] By combining the actual cancer tissue gene methylation information with the reference cancer tissue gene methylation information, and calculating the cancer tissue gene methylation information of historical patients through the ratio of the two, this process can effectively eliminate the interference of individual differences and background noise, ensuring that the obtained cancer tissue gene methylation information has higher biological relevance and predictive value. By using the average expression value of patients without pain events as a reference, the specific expression changes of cancer tissue gene methylation information in pain events can be accurately measured, thereby providing more reliable and robust data support for the prediction of acute postoperative pain risk, effectively improving the accuracy and clinical applicability of the model.

[0115] It should be noted that if Figure 8 As shown, in step 42, the process of querying the database to obtain reference patients based on the initial screening biomarker information of historical patients includes:

[0116] Step 421, obtaining basic similarity: calculating the similarity between the value of the pain assessment index of each historical patient and the value of the basic feature of other historical patients, and obtaining the basic similarity;

[0117] Step 422, obtaining the similarity of pain management plans: calculating the similarity between the pain management plan of each historical patient and the pain management plans of other historical patients respectively, to obtain the similarity of pain management plans;

[0118] Step 423, obtaining the similarity of the pain assessment index: calculating the similarity between the value of the pain assessment index of each historical patient and the value of the assessment index of other historical patients, and obtaining the similarity of the pain assessment index;

[0119] Step 424, obtaining fusion similarity: linearly weighting the basic similarity, pain management indicator similarity and pain management plan similarity, and obtaining the fusion similarity between each historical patient and other historical patients based on the values ​​corrected by the historical patient's physiological monitoring parameters, wherein the weight of the basic similarity and the weight of the pain management plan similarity are greater than the weight of the indicator similarity, and the pain assessment indicator is obtained through the patient's report results. The patient records the pain data through the smart device and shares it between the patient and the hospital.

[0120] By calculating the basic similarity, pain management scheme similarity and pain assessment index similarity, and performing linear weighting and individual feature correction, the process of obtaining fusion similarity can accurately screen out a reference patient group that is highly similar to the target patient, thereby effectively eliminating the noise caused by individual differences and ensuring that the selected reference patient data has higher relevance and reliability. This multi-dimensional similarity calculation and weight adjustment method not only comprehensively considers the comprehensive impact of basic characteristics, management schemes and assessment indicators on pain outcomes, but also enhances the sensitivity to specific biomarker information, providing strong technical support for the construction of an accurate acute postoperative pain risk prediction model.

[0121] It should be noted that the fusion similarity correction formula is:

[0122] S final (ab) = ω 1 S basic (a,b)+ω 2 S man (a,b)+ω 3 S pain (a,b)+Δ adj (a,b)

[0123] Where: a, b are the patient's indexes, and a≠b, S final (a,b) is the fusion similarity between patient a and patient b, S basic (a, b) is the basic similarity between patient a and patient b, which is calculated by cosine similarity based on the patient’s age, gender, and BMI. man (a, b) is the similarity of pain management plans between patients a and b, obtained based on the Hamming distance of pain medication, dosage, and route of administration. S pain (a, b) is the similarity of pain evaluation index between patients a and b, obtained by comparing the time-varying pain data through dynamic time warping, Δ adj (a, b) are correction items, which are corrected based on the patient's physiological monitoring data, using the mean of the difference in blood pressure, heart rate and respiratory rate as the correction item.

[0124] It should be noted that, in step 41, the formula for obtaining the actual cancer tissue gene methylation information is:

[0125]

[0126] Where: E actual is the actual cancer tissue gene methylation information, K 24 , k 48 The gene methylation information of cancer tissues at 24 hours and 48 hours after surgery was obtained by gene expression profiling (rapid PCR technology).24 , H 48 The gene methylation information of cancer tissues were respectively 24 hours and 48 hours after surgery.

[0127] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0128] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A biomarker-based method for predicting the risk of acute postoperative pain in gastric cancer patients, characterized in that: The steps include: Step 1, obtaining the initial screening biomarker information: obtaining the historical patient's initial biomarker characteristic values ​​and pain event outcomes from the hospital's historical pain management records of all gastric cancer patients, obtaining the initial screening biomarker information, and performing feature screening on the initial biomarker characteristics to obtain the target biomarker characteristics, wherein the initial biomarker characteristics include basic characteristics and pain management characteristics, the basic characteristics include clinical monitoring data characteristics, past medical history data characteristics, and drug use history data characteristics, the pain management characteristics include the patient's pain assessment index and pain management plan at the time of admission, and the pain event outcomes include whether the patient has experienced an acute postoperative pain event or not within the observation time window; Step 2, obtaining the second screening biomarker information: screening the target biomarker characteristic values ​​and pain event outcomes of each historical patient from the primary screening biomarker information to obtain the second screening biomarker information; Step 3, training an acute postoperative pain risk prediction model: using the secondary screening biomarker information to train an acute postoperative pain risk prediction model to obtain an initial target prediction model; Step 4, obtaining target biomarker information: obtaining biomarker information of the target patient, and obtaining target biomarker information according to the target biomarker characteristics; Step 5, obtaining the target pain event outcome of the target patient: input the target biomarker information into the initial target prediction model to obtain the target pain event outcome of the target patient.

2. The method for predicting the risk of acute postoperative pain in gastric cancer patients based on biomarkers according to claim 1, characterized in that: In step 1, the process of feature screening for the initial biomarker signature includes: Step 11, splitting the unit stage and the feature screening sub-stage: splitting the screening process of the initial screening biomarker information into the unit stage, and the unit stage includes the feature screening sub-stage; Step 12, obtaining a candidate screening process: obtaining a preset screening method for each unit stage, the number of preset screening methods corresponding to the feature screening sub-stage is at least 2, and the preset screening methods of all unit stages are combined to obtain a candidate screening process; Step 13, obtaining the target biomarker characteristics of the target patient: processing the initial screening biomarker information according to each candidate screening process to obtain the corresponding target biomarker characteristics.

3. The method for predicting the risk of acute postoperative pain in gastric cancer patients based on biomarkers according to claim 1, characterized in that: In step 3, the process of training the acute postoperative pain risk prediction model using the secondary screening biomarker information includes: Step 31, establishing an acute postoperative pain risk prediction model: establishing an acute postoperative pain risk prediction model based on the target biomarker characteristics corresponding to the second screening biomarker information and a preset machine learning model; Step 32, obtaining an initial candidate prediction model: training each acute postoperative pain risk prediction model according to the second screening biomarker information to obtain an initial candidate prediction model; Step 33, obtaining an initial target prediction model: evaluating the performance indicators of the initial candidate prediction models, and obtaining the initial target prediction model according to the initial candidate prediction model that best corresponds to the performance indicator.

4. The method for predicting the risk of acute postoperative pain in gastric cancer patients based on biomarkers according to claim 3, characterized in that: In step 33, the process of obtaining the initial target prediction model according to the initial candidate prediction model that best corresponds to the performance indicator includes: Step 331, obtaining a second optimal target prediction model: taking the initial candidate prediction model with the best performance index as the second optimal target prediction model; Step 332, obtaining target biomarker features according to the input features of the second best target prediction model: determining whether the number of input features of the second best target prediction model is greater than a first preset threshold, if the number of input features of the second best target prediction model is greater than the first preset threshold, arranging the input features of the second best target prediction model in descending order according to the importance of the biomarker features, and taking the biomarker features of all input features of the second best target prediction model whose ranking is less than or equal to the first preset threshold as the target biomarker features; Step 333, obtaining a third screening biomarker feature: screening the second screening biomarker information input into the second optimal target prediction model according to the target biomarker feature to obtain third screening biomarker information; Step 334, obtaining a second candidate prediction model: establishing an acute postoperative pain risk prediction model based on the three-screening biomarker information and a preset machine learning model, and using the three-screening biomarker information to train each acute postoperative pain risk prediction model to obtain a second candidate prediction model; Step 335, obtaining an initial target prediction model: evaluating the performance indicators of all second candidate prediction models, and taking the second candidate prediction model with the best performance indicator as the initial target prediction model.

5. The method for predicting the risk of acute postoperative pain in gastric cancer patients based on biomarkers according to claim 4, characterized in that: In step 332, the importance of the biomarker feature includes the degree of relevance to the outcome of the pain event and the degree of predictive contribution.

6. The method for predicting the risk of acute postoperative pain in gastric cancer patients based on biomarkers according to claim 4, characterized in that: In step 322, the process of arranging the inputs of the second-best target prediction model in descending order according to the importance of the biomarker features includes: Step 3321, obtaining the correlation degree of the pain event outcome: obtaining the correlation degree between each input feature of the two-optimal target prediction model and the pain event outcome to obtain the correlation degree of the pain event outcome, wherein the correlation degree between the input feature and the pain event outcome is obtained according to the preset screening method corresponding to the two-optimal target prediction model in the feature screening sub-stage; Step 3322, obtaining the prediction contribution degree: respectively calculating the contribution degree of each input feature of the two-optimal target prediction model to the prediction operation completed by the two-optimal target prediction model, and obtaining the prediction contribution degree; Step 3323, obtaining the importance of the input feature: the importance of the input feature is calculated according to the relevance degree of the pain event outcome and the prediction contribution degree obtained in steps 3321 and 3322 respectively.

7. The method for predicting the risk of acute postoperative pain in gastric cancer patients based on biomarkers according to claim 5, characterized in that: In step 334, the process of constructing the second candidate prediction model includes: Step 3341, obtaining four-screening biomarker information: obtaining the value of the third biomarker characteristic of the gastric cancer patient in the historical medical records, and obtaining the four-screening biomarker information; Step 3342, obtaining input features of the second candidate prediction model: using the third biomarker feature and the biomarker features corresponding to the three screening biomarker information as input features of the second candidate prediction model; Step 3343, obtaining a second candidate prediction model: using the four-screening biomarker information and the three-screening biomarker information to train an acute postoperative pain risk prediction model to obtain a second candidate prediction model.

8. The method for predicting the risk of acute postoperative pain in gastric cancer patients based on biomarkers according to claim 7, characterized in that: The third biomarker feature includes the gene methylation information of cancer tissues, which is calculated based on the promoter methylation values ​​of ALOX12, EPAS1, FANCF, HIF1A, IL10, KDM1A, KLF5, LEPR, MTAP, PTPN11, PRMT6, RBL2, and TNFAIP3 of gastric cancer patients in the pain management history records of gastric cancer patients in the hospital, and the cancer tissue methylation values ​​of CXCL12, EBF3, GPX3, HHIP, HOXA9, and TLR2 promoters; Promoter methylation values ​​of ALOX12, EPAS1, FANCF, HIF1A, IL10, KDM1A, KLF5, LEPR, MTAP, PTPN11, PRMT6, RBL2, and TNFAIP3 in cancer tissues obtained by including but not limited to tissue examination at 24 hours and 48 hours after surgery in historical gastric cancer patients, obtained by gene expression profiling; The cancer tissue methylation values ​​of CXCL12, EBF3, GPX3, HHIP, HOXA9, and TLR2 promoters include the cancer tissue methylation values ​​of CXCL12, EBF3, GPX3, HHIP, HOXA9, and TLR2 promoters in cancer tissues of historical gastric cancer patients obtained by including but not limited to tissue examination 24 hours and 48 hours after surgery.

9. The method for predicting the risk of acute postoperative pain in gastric cancer patients based on biomarkers according to claim 8, characterized in that: The process of obtaining gene methylation information of cancer tissues of gastric cancer patients in history records of pain management of gastric cancer patients in the hospital is calculated based on the promoter methylation values ​​of ALOX12, EPAS1, FANCF, HIF1A, IL10, KDM1A, KLF5, LEPR, MTAP, PTPN11, PRMT6, RBL2, and TNFAIP3 of gastric cancer patients and the cancer tissue methylation values ​​of CXCL12, EBF3, GPX3, HHIP, HOXA9, and TLR2 promoters. The process includes: Step 41, obtaining actual cancer tissue gene methylation information: according to the promoter methylation values ​​of ALOX12, EPAS1, FANCF, HIF1A, IL10, KDM1A, KLF5, LEPR, MTAP, PTPN11, PRMT6, RBL2, and TNFAIP3 of historical gastric cancer patients and the cancer tissue methylation values ​​of the promoters of CXCL12, EBF3, GPX3, HHIP, HOXA9, and TLR2, the cancer tissue gene methylation information of the corresponding historical patients is calculated; Step 42, obtaining reference cancer tissue gene methylation information: querying a database according to the initial screening biomarker information of historical patients to obtain reference patients, the reference patients being historical patients whose initial screening biomarker information similarity is greater than a second preset threshold among all historical patients who did not experience acute postoperative pain events within the observation time window, calculating the mean of the actual cancer tissue gene methylation information of all reference patients, and using the mean of the actual cancer tissue gene methylation information of all reference patients as the reference cancer tissue gene methylation information of the historical patients; Step 43, obtaining the cancer tissue gene methylation information of the historical patient: calculating the ratio of the actual cancer tissue gene methylation information to the reference cancer tissue gene methylation information, and using the ratio as the cancer tissue gene methylation information of the historical patient.

10. The method for predicting the risk of acute postoperative pain in gastric cancer patients based on biomarkers according to claim 9, characterized in that: In step 42, the process of querying the database to obtain reference patients based on the initial screening biomarker information of historical patients includes: Step 421, obtaining basic similarity: calculating the similarity between the value of the pain assessment index of each historical patient and the value of the basic feature of other historical patients, and obtaining the basic similarity; Step 422, obtaining the similarity of pain management plans: calculating the similarity between the pain management plan of each historical patient and the pain management plans of other historical patients respectively, to obtain the similarity of pain management plans; Step 423, obtaining the similarity of the pain assessment index: calculating the similarity between the value of the pain assessment index of each historical patient and the value of the assessment index of other historical patients, and obtaining the similarity of the pain assessment index; Step 424, obtaining fusion similarity: linearly weighting the basic similarity, pain management indicator similarity and pain management plan similarity, and obtaining the fusion similarity between each historical patient and other historical patients based on the values ​​corrected by the historical patient's physiological monitoring parameters, wherein the weight of the basic similarity and the weight of the pain management plan similarity are greater than the weight of the indicator similarity, and the pain assessment indicator is obtained through the patient's report results. The patient records the pain data through the smart device and shares it between the patient and the hospital.

Images