Metabolic marker combination for predicting postoperative nausea and vomiting risk and application thereof

Abstract

The invention relates to the technical field of biomarkers. The invention provides a metabolic marker combination for predicting the risk of postoperative nausea and vomiting and application of the metabolic marker combination. The metabolic marker combination comprises isodeoxycholic acid, 20-hydroxyprostaglandin F2alpha, carnitine C9: 1, N-methyltryptamine and L-leucine-L-leucine-L-glycine. 0.9) is shown in a training set and a verification set by a discrimination model constructed by the metabolic marker combination, and the discrimination model has good mechanism interpretability and is suitable for postoperative individual risk assessment. The metabolite combination and the distinguishing method thereof can be widely applied to perioperative risk management, clinical aid decision making and individualized anesthesia scheme optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of biomarkers, and in particular to a metabolic marker combination for predicting the risk of postoperative nausea and vomiting and applications thereof. Background Art

[0002] Postoperative nausea and vomiting (PONV) is a common perioperative clinical complication. The incidence of nausea after general anesthesia is approximately 50%, and the incidence of vomiting is approximately 30%. In some individuals at high risk for PONV, the incidence can reach 80%. Besides causing intense subjective discomfort, persistent nausea and vomiting can also lead to serious complications such as dehydration, electrolyte and acid-base imbalances, esophageal lacerations, aspiration and pneumonia, subcutaneous emphysema, increased abdominal pressure leading to wound dehiscence and bleeding, increased wound pain, and even asphyxiation and death. Therefore, PONV significantly impacts postoperative recovery, prolongs postoperative hospital stays, increases hospital medical costs, and significantly reduces patient comfort and surgical satisfaction.

[0003] The incidence of PONV varies from person to person, and different levels of preventive measures are required for patients with different risk factors. The most commonly used method for PONV risk assessment in adults is the simplified risk assessment system summarized by Apfel et al. In addition, the fourth edition of the guidelines for the prevention and treatment of PONV released by the International Society for Anesthesia Research in May 2020 recommends that for people at high risk of PONV, combination medication can be used to prevent and treat nausea and vomiting. Although multi-drug combination therapy for preventing postoperative nausea and vomiting (PONV) is a common clinical practice, it also has some potential disadvantages: for example, the use of multiple drugs may increase the risk of drug interactions, which may lead to weakened or enhanced drug efficacy or even adverse reactions. At the same time, certain drug combinations may affect drug metabolism, thereby changing its blood concentration, which may require additional monitoring and adjustment. In addition, the combined use of multiple drugs may also increase the chances of patients experiencing side effects such as dizziness, headache, constipation or dry mouth. The side effects of different drugs may overlap, causing increased discomfort for patients and affecting the quality of their postoperative recovery.

[0004] Therefore, preoperative prediction of PONV is crucial. Accurately predicting PONV can help physicians develop more effective prevention strategies and tailor preventive and treatment options to the patient's specific circumstances. This can reduce postoperative discomfort and improve the patient's overall postoperative experience and satisfaction, while also avoiding one-size-fits-all treatment options and reducing unnecessary medication use and associated side effects.

[0005] Currently, the risk of PONV is primarily predicted using a simplified risk assessment system developed by Apfel et al. However, its predictive performance is significantly limited, relying solely on a limited number of clinical risk factors, such as gender, smoking history, prior history of PONV or motion sickness, and postoperative opioid use. It fails to account for other PONV factors, such as genetics, related to drug metabolism, the role of neurotransmitters, and pain sensitivity, and fails to fully account for inter-individual physiological and metabolic differences. However, metabolomics analysis of surgical patient blood samples using high-performance liquid chromatography coupled with quadrupole time-of-flight mass spectrometry can provide more comprehensive biomarker information, including comprehensive insights into multiple metabolic pathways and biological processes associated with PONV. This provides more comprehensive biomarker information, as well as individual metabolic signatures, and provides more precise markers for personalized PONV prediction. Summary of the Invention

[0006] The present invention aims to provide a combination of metabolite markers for predicting the risk of postoperative nausea and vomiting (PONV), and its application. By performing metabolomics analysis on preoperative blood samples and using elastic net and random forest methods, seven metabolites significantly associated with PONV were identified. Further, combined with mechanistic studies, five metabolites with clear roles in physiological regulation were incorporated to construct a predictive model with excellent performance and biological mechanism support. This model enables personalized assessment of preoperative PONV risk and is suitable for a variety of clinical scenarios, including anesthesia pre-assessment, postoperative rehabilitation pathway design, and precision intervention.

[0007] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions: The present invention provides a metabolic marker combination for predicting the risk of postoperative nausea and vomiting, wherein the metabolic marker combination includes isodeoxycholic acid, 20-hydroxyprostaglandin F2α, carnitine C9:1, N-methyltryptamine, and L-leucine-L-leucine-L-glycine; The IUPAC name of the carnitine C9:1 is: 3-[(E)-non-2-enoyl]oxy-4-(trimethylazaniumyl)butanoate.

[0008] Preferably, the carnitine C9:1 has been included in the PubChem database, see the link for details: https: / / pubchem.ncbi.nlm.nih.gov / #query=3-%5B(E)-non-2-enoyl%5Doxy-4-(trimethylazaniumyl)butanoate.

[0009] Preferably, the metabolic marker combination further includes one or more of palmitolamine, androstenediol, 3-hydroxy-L-phenylalanine, and serotonin.

[0010] The present invention also provides the use of the metabolic marker combination as a risk marker for predicting postoperative nausea and vomiting.

[0011] Preferably, the subject of the application is a patient after general anesthesia.

[0012] Preferably, the patient is a laparoscopic gastrointestinal surgery patient.

[0013] The present invention also provides the use of the metabolic marker combination in preparing a product for predicting the risk of postoperative nausea and vomiting.

[0014] Preferably, the product is a kit.

[0015] The present invention also provides a kit for predicting the risk of postoperative nausea and vomiting, which comprises the metabolic marker combination, a reagent for detecting the metabolic marker combination, and software or an algorithm module for generating a PONV risk score based on the detection results.

[0016] The present invention also provides a method for constructing a risk model for predicting postoperative nausea and vomiting using the above-mentioned combination of metabolic markers, which is characterized by comprising the following steps: (1) Collect preoperative blood samples; (2) Perform metabolomics testing to obtain metabolite expression profiles; (3) extracting metabolite expression data from the metabolite marker combination; (4) Input the above data into the machine learning algorithm to build a prediction model.

[0017] Preferably, the machine learning algorithm is elastic net regression or random forest algorithm.

[0018] The present invention provides a metabolite marker combination for predicting the risk of postoperative nausea and vomiting, and its application. The metabolite marker combination includes isodeoxycholic acid, 20-hydroxyprostaglandin F2α, carnitine C9:1, N-methyltryptamine, and L-leucine-L-leucine-L-glycine. The discriminant model constructed using this metabolite marker combination demonstrated high predictive accuracy (AUC > 0.9) in both training and validation sets, and exhibited good mechanistic explainability, making it suitable for personalized postoperative risk assessment. This metabolite combination and its discriminant method can be widely used in perioperative risk management, clinical decision support, and personalized anesthesia regimen optimization. DETAILED DESCRIPTION

[0019] The technical solutions provided by the present invention are described in detail below with reference to the embodiments, but they should not be construed as limiting the scope of protection of the present invention.

[0020] Example 1 Metabolome Data Collection 1. Collecting Plasma Samples The present invention first obtained preoperative plasma from 32 adult patients at high risk of PONV who did not vomit postoperatively and 32 adult patients who vomited postoperatively and underwent laparoscopic surgery through a clinical research project (clinical research registration number: NCT04853147). The plasma was obtained by the following steps: 1. Collect 3 ml of preoperative whole blood using a sodium citrate anticoagulant blood collection tube; Centrifuge at 2.500 g for 10 minutes, and collect the upper plasma layer after the blood cells are precipitated; 3. Collect the supernatant of plasma and centrifuge it at 2000g for 15 minutes to obtain the plasma sample after removing the precipitate; 4. Aliquot and freeze in a -80℃ freezer.

[0021] 2. Sample Extraction Process 1. Take out the sample from the -80℃ freezer and thaw it on ice until there is no ice in the sample; 2. After the sample is thawed, vortex for 10 seconds to mix and transfer 50 μl of sample; 3. Add 300 μl of 20% acetonitrile methanol internal standard extract, vortex for 3 minutes, and centrifuge at 12,000 rpm for 10 minutes at 4°C. 4. After centrifugation, transfer 200 μl of the supernatant to another corresponding numbered centrifuge tube and place it in a -20°C refrigerator for 30 minutes; 5. Centrifuge at 12,000 rpm for 3 minutes at 4°C. Transfer 180 μl of the supernatant to the corresponding injection bottle liner for analysis.

[0022] 3. Chromatographic mass spectrometry acquisition conditions 1. T3 chromatographic conditions (1) Chromatographic column: Waters ACQUITY UPLC HSS T3 C18 1.8um, 2.1mm*100mm; (2) Mobile phase: Phase A is ultrapure water (0.1% formic acid), and phase B is acetonitrile (0.1% formic acid); (3) Flow rate: 0.4 ml / min; column temperature 40°C; injection volume: 2 μl (broadly targeted detection), 5 μl (non-targeted detection).

[0023] Table 1. Gradient conditions of positive and negative ion mobile phases for T3 chromatographic column Time (min) A（%） B（%） 0.0 95 5 2.0 80 20 5.0 40 60 6.0 1 99 7.5 1 99 7.6 95 5 10.0 95 5 2. HILIC Chromatographic Conditions (1) Chromatographic column: Waters ACQUITY UPLC BEH HILIC Column, 1.7 µm, 1 mm * 100 mm; (2) Mobile phase A: 20 mM ammonium formate, 30% water, 10% methanol, 60% acetonitrile, adjusted to pH 10.6 with aqueous ammonia; (3) Instrument column temperature: 40°C; flow rate: 0.4 ml / min; injection volume: 2 μl (broad targeted detection), 5 μl (non-targeted detection).

[0024] Table 2 HILIC column mobile phase gradient conditions Time (min) A（%） B（%） 0.00 95 5 3.50 70 30 5.50 5 95 6.50 5 95 6.51 95 5 10.00 95 5 3. Non-targeted mass spectrometry conditions The data acquisition instrument system mainly includes Ultra Performance Liquid Chromatography (UPLC) and Quadrupole-Time of Flight mass spectrometry Table 3 AB Triple TOF 6600 mass spectrometry conditions Chinese name ESI+ ESI- Acquisition time (min) 10 10 Ionization voltage (V) 5000 -4000 Ion source temperature (℃) 550 450 Spray gas (PSI) 50 50 Auxiliary heating gas (PSI) 60 60 Curtain Air (PSI) 35 35 Declustering voltage (V) 60 -60 MSI collision energy (V) 10 -10 MS2 collision energy (V) 30 -30 Collision energy step (V) 15 15 4. Collection conditions for broad targeted testing Data acquisition instruments include ultra-high performance liquid chromatography (UPLC) and tandem mass spectrometry (MS / MS) Table 4 AB QTRAP® 6500 mass spectrometry conditions Chinese name ESI+ ESI- Acquisition time (min) 10 10 Ionization voltage (V) 550 -4500 Ion source temperature (℃) 50 500 Spray gas (PSI) 50 50 Auxiliary heating gas (PSI) 50 50 Curtain Air (PSI) 25 25 Collision-induced ionization high high Principles of Metabolite Qualitative and Quantitative Identification All samples were extracted and mixed in equal amounts to form QC samples, and non-targeted detection was performed on the LC-QTOF-MS / MS platform. Accurate identification was performed based on the self-built standard database MWDB (including secondary spectra and retention time RT), Maiwei's integrated DB-all public database (including Metlin, HMDB, KEGG and other databases) AI prediction library and MetDNA. Multiple ion pairs and retention time RT of identified metabolites were extracted, and combined with Maiwei's self-built target database to form a new library exclusive to the project. Finally, in all samples, MRM precise quantification of metabolites in the new library was performed based on the Q-Trap instrument platform.

[0025] Implementation Case 2 Candidate Marker Screening Process 1. Data Preprocessing The sample size for this project was 64, and the number of metabolites used was 1207. Due to the large variability of the metabolite quantitative data, the data were first log2-transformed. Individuals were divided into two groups based on whether they vomited or did not vomit within 24 hours after surgery. Each combination was randomly split into a training set and a validation set at a ratio of 2:1, resulting in a total of 42 individuals in the training set and 22 individuals in the validation set, respectively.

[0026] 2. Machine Learning to Select Metabolites To further narrow the pool of biomarker candidates, variable selection methods are needed to screen metabolites. Variable selection involves using statistical analysis to select a set of independent variables with strong explanatory power for the dependent variable from a large pool of independent variables. Here, we use the elastic net method and random forest (RF) to select metabolites, which are then combined to generate candidate biomarkers.

[0027] (1) Using Elastic Net to select metabolites The elastic net method combines LASSO and ridge regression, allowing groups of highly correlated variables to be included in the model together. This improves on the shortcomings of LASSO while also incorporating the advantages of ridge regression in handling collinearity. The elastic net method is implemented using the R package glmnet.

[0028] This project ran glmnet 500 times, randomly sampling 80% of the training set without replacement each time for modeling. Finally, metabolites with a frequency of 400 or more were selected. The following table shows the metabolites selected using the elastic net method.

[0029] Table 5 serial number substance Substance Category 1 Isodeoxycholic acid Bile acids 2 20-hydroxyprostaglandin F2α Fatty acyl groups 3 N-phenylanthranilic acid Benzene and its derivatives 4 N-methyltryptamine Tryptamine, choline, pigment 5 Carnitine C9:1 Fatty acyl groups 6 L-Leucine-L-Leucine-L-Glycine Amino acids and their metabolites 7 (S)-2-Methyl-3,4,5,6-tetrahydropyrimidine-4-carboxylic acid Heterocyclic compounds 8 Carboxytolbutamide Benzene and its derivatives 9 4-Hydroxy-3-(3-methylbut-2-enyl)benzoic acid Benzene and its derivatives 10 - Benzene and its derivatives 11 Acetosyringone Aldehydes, ketones, esters 12 3-(3-Hydroxy-4-methoxyphenyl)propionic acid Organic acids and their derivatives 13 5-[6-Hydroxy-2-(4-hydroxyphenyl)-4-[(E)-2-(4-hydroxyphenyl)vinyl]-1-benzofuran-3-yl]benzene-1,3-diol Benzene and its derivatives 14 epicatechin flavonoids 15 Methionylserinamide Amino acids and their metabolites (2) Using random forest to select metabolites Random forests are an ensemble type of machine learning algorithm. They use bootstrap aggregating to extract multiple samples with replacement from the original sample as a training set. They then build a decision tree model on the training set, combine the predictions from multiple decision trees, and vote to arrive at the final prediction. RFs offer high prediction accuracy, are highly tolerant to outliers and noise, and are less susceptible to overfitting.

[0030] When the training set is extracted using the bootstrap clustering method, the probability that a sample in the original data is not extracted is (1–1 / N) N, where N is the number of samples in the original data. When N is large enough, (1–1 / N) N The convergence rate is 1 / e≈0.368. This indicates that nearly 37% of the samples in the original data do not appear in the extracted training set. This data is called out-of-bag (OOB) data. This OOB data is used to estimate model performance. Random forest is implemented using the R package varSelRF.

[0031] Table 6 serial number substance Substance Category 1 N-phenylacetylglycine Amino acids and their metabolites 2 1-naphthol Benzene and its derivatives 3 Isodeoxycholic acid Bile acids 4 20-hydroxyprostaglandin F2α Fatty acyl groups 5 N-phenylanthranilic acid Benzene and its derivatives 6 N-methyltryptamine Tryptamine, choline, pigment 7 Carnitine C9:1 Fatty acyl groups 8 L-Leucine-L-Leucine-L-Glycine Amino acids and their metabolites 9 (S)-2-Methyl-3,4,5,6-tetrahydropyrimidine-4-carboxylic acid Heterocyclic compounds 10 Carboxytolbutamide Benzene and its derivatives 11 4-Hydroxy-3-(3-methylbut-2-enyl)benzoic acid Benzene and its derivatives 12 - Benzene and its derivatives 13 Cyclohexane other 14 - Heterocyclic compounds 15 Acetosyringone Aldehydes, ketones, esters 16 3-(phenylsulfonyl)-1-(4-fluorophenyl)pyrrolo[3,2-b]quinoxalin-2-amine Benzene and its derivatives 17 Lonidam Heterocyclic compounds 18 3-(3-Hydroxy-4-methoxyphenyl)propionic acid Organic acids and their derivatives 19 5-[6-Hydroxy-2-(4-hydroxyphenyl)-4-[(E)-2-(4-hydroxyphenyl)vinyl]-1-benzofuran-3-yl]benzene-1,3-diol Benzene and its derivatives 20 epicatechin flavonoids 21 Methionylserinamide Amino acids and their metabolites 22 1-stearoyl-2-hydroxy-sn-glycero-3-phosphoethanolamine other (3) Selecting the best biomarker Since each model has certain shortcomings, it is necessary to integrate the metabolites selected by the elastic net and random forest algorithms, and then use the validation set samples to evaluate the effectiveness of the metabolites. The intersection of the metabolites selected by the two algorithms is as follows: Table 7 serial number substance Substance Category 1 Isodeoxycholic acid Bile acids 2 20-hydroxyprostaglandin F2α Fatty acyl groups 3 N-phenylanthranilic acid Benzene and its derivatives 4 N-methyltryptamine Tryptamine, choline, pigment 5 Carnitine C9:1 Fatty acyl groups 6 L-Leucine-L-Leucine-L-Glycine Amino acids and their metabolites 7 (S)-2-Methyl-3,4,5,6-tetrahydropyrimidine-4-carboxylic acid Heterocyclic compounds 8 Carboxytolbutamide Benzene and its derivatives 9 4-Hydroxy-3-(3-methylbut-2-enyl)benzoic acid Benzene and its derivatives 10 organic phosphites Benzene and its derivatives 11 Acetosyringone Aldehydes, ketones, esters 12 3-(3-Hydroxy-4-methoxyphenyl)propionic acid Organic acids and their derivatives 13 5-[6-Hydroxy-2-(4-hydroxyphenyl)-4-[(E)-2-(4-hydroxyphenyl)vinyl]-1-benzofuran-3-yl]benzene-1,3-diol Benzene and its derivatives 14 epicatechin flavonoids 15 Methionylserinamide Amino acids and their metabolites 3. Comprehensive screening and mechanism strengthening: Considering the potential for false positives or lack of biological interpretability in machine learning methods, and to enhance the model's biological plausibility and clinical interpretability, combined with our understanding of the pathogenesis of PONV, we further screened five candidate metabolites that have been closely associated with the pathogenesis of postoperative vomiting in relevant literature and research. These metabolites include isodeoxycholic acid, 20-hydroxyprostaglandin F2α, carnitine C9:1, N-methyltryptamine, L-leucine-L-leucine-L-glycine, and the mechanism-enhancing metabolites palmitoylethanolamine, androstenediol, 3-hydroxy-L-phenylalanine, and serotonin. Although these metabolites individually had limited predictive power (AUC < 0.9) in this project's data, their physiological functions are highly relevant to the pathogenesis of PONV and were therefore included in the model as a mechanism-enhancing group.

[0032] Example 3 Final combination and modeling This example is based on the 9 candidate metabolites obtained in the final screening 1. Data Preprocessing: The following preprocessing steps were performed on the nine metabolite features identified: Log2 transformation: All metabolite concentration data were log2 transformed to compress the numerical range; Z-score normalization: The log2-transformed variables were further normalized to a mean of 0 and a standard deviation of 1 to reduce the impact of scale differences between variables on model formation. After these preprocessing steps, a feature matrix was generated that could be directly used for modeling and analysis.

[0033] 2. Model Construction and Validation: Logistic regression was used to model the standardized feature matrix. The samples were randomly divided into a training set (n=42) and a validation set (n=22) in a 2:1 ratio. A logistic regression model was fitted to the training set to predict the probability of an event. Model performance was evaluated in the validation set, with the area under the receiver operating characteristic (ROC) curve (AUC) reaching 1.00 and a prediction accuracy of 100%, demonstrating good discriminatory power. The model coefficients are as follows: Table 8 serial number name coefficient 1 Isodeoxycholic acid 0.801948 2 20-hydroxyprostaglandin F2α 0.92907 3 Carnitine C9:1 0.929886 4 N-methyltryptamine -0.71375 5 L-Leucine-L-Leucine-L-Glycine -0.80326 6 Palm Ethanolamine 0.074023 7 androstenediol 0.114469 8 3-Hydroxy-L-phenylalanine 0.054092 9 serotonin 0.103151 The model intercept term is +0.0520.

[0034] 3. Prediction Model: The above logistic regression coefficients are used to construct the prediction formula as follows:

[0035] Among them, x1~x9 are standardized variables and β is the model coefficient.

[0036] Among them, xix_ixi is the Z-score standardized value of the iii-th metabolite, βi\beta_iβi is the corresponding logistic regression coefficient, and β0\beta_0β0 ​​is the intercept term.

[0037] Example 4 Cross-validation and performance evaluation strategy of prediction model

[0038] This embodiment provides a cross-validation method for evaluating the stability and generalization ability of a prediction model, ensuring that the model has stable prediction performance under different sample combinations and avoiding overfitting due to sample bias.

[0039] 1. Cross-validation method: 5-fold Stratified K-Fold cross-validation method is used: (1) Divide all samples into 5 parts, using 4 parts for training and 1 part for validation each time; (2) Repeat the training-validation process for 5 rounds to obtain the prediction results for each round; (3) Comprehensively evaluate the performance stability of the model under different subsets.

[0040] Performance evaluation indicators: After cross-validation, the predicted probability of each sample is integrated, the ROC curve is drawn, and the AUC is calculated as follows: (1) Cross-validation AUC: 1.00; (2) Average accuracy: 100%; (3) The sensitivity and specificity were both 100%.

[0041] Example 5 Prediction Model Deployment System and Software Interface This embodiment provides a method for deploying a prediction tool based on a constructed logistic regression model, including two forms: an Excel template and a web calculator. Users can directly input metabolite test values ​​without the need for professional modeling software (such as Python or SPSS) to obtain risk prediction results for postoperative complications.

[0042] In order to make the model ready for use in clinical or scientific research sites, the present invention embeds the logistic regression model into an Excel spreadsheet and implements prediction operations through function formulas.

[0043] (1) Input area: Columns A to I correspond to the original detection concentrations of the 9 metabolites. The user needs to enter the 9 metabolite values ​​of the sample to be tested into the corresponding cells (no standardization is required).

[0044] (2) Operation area: The J column contains built-in prediction formulas, which implement log2 transformation + Z standardization + logistic regression probability calculation; (3) Output: Column J returns the predicted value, which is the probability value of the event (range 0 to 1). The threshold value (such as 0.5) can be used to further determine whether it is a high risk; (4) Template advantages: No software installation is required and it can be opened and used on any computer; it supports batch input and batch calculation; it can be exported as a report for electronic medical records or scientific research analysis.

[0045] As can be seen from the above examples, the present invention provides a metabolite marker combination for predicting the risk of postoperative nausea and vomiting, and its application. The metabolite marker combination includes isodeoxycholic acid, 20-hydroxyprostaglandin F2α, carnitine C9:1, N-methyltryptamine, and L-leucine-L-leucine-L-glycine. The discriminant model constructed using this metabolite marker combination demonstrated high predictive accuracy (AUC > 0.9) in both training and validation sets, and possessed good mechanistic explainability, making it suitable for personalized postoperative risk assessment. This metabolite combination and its discriminant method can be widely applied in perioperative risk management, clinical decision support, and personalized anesthesia regimen optimization.

[0046] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A metabolic marker combination for predicting the risk of postoperative nausea and vomiting, characterized in that: The metabolic marker combination includes isodeoxycholic acid, 20-hydroxyprostaglandin F2α, carnitine C9:1, N-methyltryptamine, and L-leucine-L-leucine-L-glycine; The IUPAC name of the carnitine C9:1 is: 3-[(E)-non-2-enoyl]oxy-4-(trimethylazaniumyl)butanoate.

2. The metabolic marker combination according to claim 1, characterized in that The metabolic marker combination further includes one or more of palmitolamine, androstenediol, 3-hydroxy-L-phenylalanine, and serotonin.

3. Use of the metabolic marker combination according to claim 1 or 2 as a risk marker for predicting postoperative nausea and vomiting.

4. The use according to claim 3, characterized in that The object of the application is a patient after general anesthesia.

5. The use according to claim 4, characterized in that The patient is a laparoscopic gastrointestinal surgery patient.

6. Use of the metabolic marker combination according to claim 1 or 2 in the preparation of a product for predicting the risk of postoperative nausea and vomiting.

7. The use according to claim 6, characterized in that The product is a test kit.

8. A kit for predicting the risk of postoperative nausea and vomiting, characterized in that: The kit comprises the metabolic marker combination according to claim 1 or 2, a reagent for detecting the metabolic marker combination according to claim 1 or 2, and software or an algorithm module for generating a PONV risk score based on the detection results.

9. A method for constructing a risk model for predicting postoperative nausea and vomiting using the combination of metabolic markers according to claim 1 or 2, characterized in that: The steps include: (1) Collect preoperative blood samples; (2) Perform metabolomics testing to obtain metabolite expression profiles; (3) extracting metabolite expression data from the metabolite marker combination according to claim 1 or 2; (4) Input the above data into the machine learning algorithm to build a prediction model.

10. The method according to claim 9, characterized in that The machine learning algorithm is elastic net regression or random forest algorithm.

Images