A prediction model for metabolic associated fatty liver disease and liver disease death based on metabolomics and its uses

Abstract

The present invention discloses a metabolic associated steatohepatitis and liver disease death prediction model based on metabolomics and its uses. The model formula is as follows: MASH prediction score = 0.137 × BMI (kg / m<supgt;2< / supgt>) + 0.054 × AST (U / L) + 0.033 × Tyr (mol / L) - 0.421 × V0PLp (%) - 0.243. This model is based on NMR metabolomics and machine learning techniques to explore a new non-invasive scoring model, which is of great significance for early screening and diagnosis of metabolic associated steatohepatitis and improving the survival rate of patients.

Description

Technical Field

[0001] The present invention relates to a biological disease death prediction model and its uses, and particularly to a metabolic - associated steatohepatitis and liver disease death prediction model based on metabolomics and its uses. Background Art

[0002] In recent years, with the improvement of living standards, the incidence of metabolic - associated fatty liver disease (MASLD) has been increasing globally. Moreover, MASLD is closely related to obesity, type 2 diabetes, cardiovascular diseases, etc. Once such patients progress to the stage of metabolic - associated steatohepatitis (MASH), the incidence of liver cancer and liver - specific mortality are both 10 times higher than those of patients without MASH. Therefore, timely and accurately identifying MASH is crucial for evaluating the clinical prognosis of patients and guiding individualized treatment.

[0003] Previously, it was considered that among the existing detection techniques, the pathological result of liver tissue biopsy was the gold standard for diagnosing MASH. However, this invasive operation brings great pain to patients, and there are risks such as infection, bleeding, and damage to adjacent organs during the operation. Post - operatively, there may also be complications such as pain at the puncture site, pneumothorax or empyema, and biliary peritonitis. Moreover, the operation is complex, expensive, and has many limitations in clinical application scenarios. Among non - invasive imaging examination methods, liver ultrasound and CT can evaluate the size, density, morphology, and the state of intra - hepatic tubular structures of the liver, thereby indirectly evaluating the liver fat accumulation. However, they lack characteristic signs and have low diagnostic specificity. Liver magnetic resonance imaging has no radiation, but the examination process is complex, time - consuming, expensive, and has a limited scope of application. Fibroscan liver elastography and steatosis quantification technology are emerging diagnostic technologies in recent years. They measure the liver stiffness and fibrosis degree based on ultrasonic transient elastography, but they also have the problem of low diagnostic specificity. Positive results can also be obtained for increased liver stiffness caused by viral hepatitis, alcoholic hepatitis, etc.

[0004] Globally, MASLD has become a major cause of morbidity and mortality related to liver diseases. Therefore, clinically identifying MASLD patients at high risk of liver disease death is a key issue. Previous studies have shown that severe hepatic steatosis shown by ultrasound examination or surrogate parameters (including liver enzymes and simple steatosis indices (the Fibrosis-4 index, the Fatty Liver Index, the NAFLD Liver Fat Score, and the Hepatic Steatosis Index)) is associated with an increased risk of liver-specific mortality. However, these parameters are not particularly effective because participants in the highest decile of liver enzymes or with medium / high steatosis indices have only a 4- to 5-fold increased risk of dying from liver disease. Clinically, simple liver fibrosis scores such as FIB-4, NFS, or APRI have been widely used to predict advanced liver fibrosis, and they can reduce the need for liver biopsy. However, multiple studies have shown that these scores have at most moderate accuracy in predicting all-cause mortality (AUROCs 0.71 - 0.72) and liver-specific mortality (AUROCs 0.67 - 0.83). The accuracy of these fibrosis scores in predicting liver-specific mortality is limited because the proportion of false negatives is non-negligible, and the accuracy in detecting MASH and early fibrosis is insufficient. In fact, even histological assessment of fibrosis can only predict all-cause mortality, with an AUROC of 0.72 (0.62 - 0.81), similar to the accuracy of liver stiffness measured by vibration-controlled transient elastography, FIB-4, and NFS (AUROCs 0.70 - 0.76). In the absence of advanced fibrosis, the MASH stage is associated with a more than 10-fold increase in liver-specific mortality and is recommended as a treatment target.

[0005] Studies have found that MASH is closely related to metabolic syndrome and type 2 diabetes, often accompanied by elevated serum alanine aminotransferase (ALT), aspartate aminotransferase (AST), and inflammation and fibrosis markers. However, since these molecules can be elevated in multiple disease settings, the detection specificity and sensitivity of these biochemical markers are low. Previous studies on non-invasive diagnosis of MASH using traditional biomarkers have shown low accuracy or lack of independent validation in multiple ethnic groups. In the past two years, several international scores have combined imaging and serum biomarkers to evaluate active MASH with liver fibrosis, such as the FAST score and NIS4, but the value of these indices in predicting long-term all-cause mortality and cause-specific mortality remains to be studied. Considering the increasing incidence of MASH in the population, it is urgent to explore new diagnostic models using minimally invasive and easily accessible body fluid samples to help with the early screening of MASH.

[0006] Metabolomics studies the dynamic changes of metabolites in different tissues, cells, and body fluids of organisms after physiological and pathological stimuli such as gene mutations or environmental perturbations. As the most downstream part of systems biology, it is closest to the functional state of organisms and disease phenotypes. Among them, metabolomics based on nuclear magnetic resonance (NMR) has become a valuable technical application due to its non-invasive nature, short detection time, high reproducibility, and quantitative detection. The sample processing steps before NMR analysis are fewer, avoiding biologically irrelevant biases and interferences. In addition, NMR is a non-contact technology that can perform single-dimensional or multi-dimensional detection on a single sample, obtaining comprehensive parameter information in a short time. In the past 20 years, NMR metabolomics technology has been widely applied to the research of various disease backgrounds, and has also provided new ideas for the evaluation and diagnosis of metabolic associated steatohepatitis.

[0007] Machine learning is an artificial intelligence science that uses theories in probability theory, statistics, and complex algorithms, uses a computer as a tool, and real-time simulates the human learning method to divide the existing content into knowledge structures. It can continuously improve the performance of specific algorithms in empirical learning and has been considered to have great application potential in the future medical field. A recent study conducted in the Caucasian population found that two non-invasive MASH scores, NFS and FIB-4, can accurately predict the prognosis, and the machine learning-based method is a feasible solution for further optimizing the scoring system.

[0008] Therefore, exploring new non-invasive scoring models is of great significance for the early screening and diagnosis of metabolic associated steatohepatitis and improving the survival rate of patients. Summary of the Invention

[0009] Object of the Invention: The object of the present invention is to provide a safe, accurate, stable, fast, and highly reproducible prediction model for metabolic associated steatohepatitis and liver disease death based on metabolomics. Based on NMR metabolomics and machine learning technologies, exploring new non-invasive scoring models is of great significance for the early screening and diagnosis of metabolic associated steatohepatitis and improving the survival rate of patients.

[0010] Technical Solution: The present invention provides a prediction model for metabolic associated steatohepatitis and liver disease death based on metabolomics, and the model formula is as follows:

[0011] MASH prediction score = 0.137 × BMI (kg / m 2 ) + 0.054 × AST (U / L) + 0.033 × Tyr (mol / L) – 0.421 × V0PLp (%) – 0.243,

[0012] Among them are body mass index (BMI), serum aspartate aminotransferase (AST), tyrosine concentration, and the ratio of phospholipids to total lipids in VLDL (V0PLp).

[0013] Furthermore, the prediction model adopts a strategy combining NMR metabolomics and machine learning.

[0014] Furthermore, the detection parameters used in 1H-NMR serum metabolomics detection are as follows: at a temperature of 310K, the number of data points is 98k, the number of transients is 32, the spectral width is 20 ppm, and the line broadening factor is 0.3 Hz.

[0015] Furthermore, the detection equipment used in 1H-NMR serum metabolomics detection is a 600 MHz AVANCE III NMR spectrometer equipped with a BBI probe, and high-throughput NMR metabolomics analysis technology is used to measure serum metabolites.

[0016] Furthermore, 1H-NMR serum metabolomics detection uses the 1D water presaturation NOESY (NOESYGPPR1D) pulse sequence to detect one-dimensional NMR hydrogen spectrum signals.

[0017] Furthermore, 1H-NMR serum metabolomics detection uses the presaturated bipolar gradient longitudinal eddy current delay (LEDBPGPPR2S1D) pulse sequence to detect NMR diffusion-edited spectra, attenuating small molecule signals and retaining macromolecule signals.

[0018] Furthermore, after the 1H-NMR serum metabolomics detection is completed, in MATLAB R2017a (Mathworks Inc.) software, an algorithm based on the simplex method is used to quantify all signals to prevent the detection lower limit of specific signals from being too high due to signal overlap and high diversity.

[0019] Furthermore, after signal quantification is completed, version 3.6.0 of Topspin software is used to complete the preprocessing (baseline adjustment, phase correction, calibration) of all NMR spectra, and integration and quantitative calculations of all NMR spectrum signals are performed based on version 4.3.0 of R software.

[0020] The detection equipment is a 600 MHz AVANCE III NMR spectrometer, equipped with a BBI probe (Bruker Biospin GmbH, Germany), and high-throughput NMR metabolomics analysis services (Nightingale Health Ltd., Helsinki, Finland) are used. Version 4.3.0 of R software is used to complete the quantitative analysis of metabolites. The NMR detection method includes standard 1D 1HNMR experiments and diffusion editing experiments. The 1D NMR hydrogen spectrum signals were detected using the 1D water pre-saturated NOESY (NOESYGPPR1D) pulse sequence; the NMR diffusion editing spectrum was detected using the pre-saturated bipolar gradient longitudinal eddy current delay (LEDBPGPPR2S1D) sequence. By optimizing parameters such as the diffusion delay time, the signals of small molecules were attenuated while the signals of macromolecules were retained. All NMR spectra were pre-processed (baseline adjustment, phase correction, calibration) in the Topspin software version 3.6.0, and then based on the R software version 4.3.0, combined with the R package "rDolphin", the signals of all NMR spectra were integrated and quantitatively calculated.

[0021] Furthermore, it includes body mass index (BMI), serum glutamic-oxaloacetic transaminase (AST) and tyrosine concentration, and the ratio of phospholipids to total lipids in VLDL (V0PLp).

[0022] On the other hand, the present invention provides the use of the said model in the prediction of metabolic associated steatohepatitis and liver disease death based on metabolomics.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0024] (1) NMR has the characteristics of strong stability and good repeatability, so the individual heterogeneity of the detection results is small;

[0025] (2) By adopting multiple clinical parameters and metabolite levels, and using machine learning to optimize the model, it can more comprehensively capture various pathological changes and functional states of the liver during the MASLD course;

[0026] (3) The detection time is short, it can efficiently and simply quantify the levels of serum-related metabolites, and then early screen for MASH according to the changes in their levels;

[0027] (4) The detection process is based on blood specimens with little trauma and convenient sampling. The human body does not need to be exposed to radiation, and it has the advantages of safety, reliability, no harm to the human body, low detection cost, and small economic burden on patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 Bayesian information criterion graph for clinical variables for best subset regression screening;

[0029] Figure 2 Statistics of four candidate variables, namely BMI, AST, Tyr and V0PLp, for MASH patients and non-MASH subjects in the Chinese liver biopsy cohort;

[0030] Figure 3 Correlation between the MASH prediction score and hepatic steatosis, ballooning, lobular inflammation and fibrosis scores in the Chinese liver biopsy cohort;

[0031] Figure 4 AUROC of the MASH prediction score in the liver biopsy cohorts in China and Finland;

[0032] Figure 5 Sensitivity and specificity of the MASH prediction score in the liver biopsy cohorts in China and Finland;

[0033] Figure 6 Liver-specific mortality and NAFLD / MASLD-related mortality in high-risk and low-risk populations among all participants in the Shanghai Changfeng cohort and the UK Biobank cohort;

[0034] Figure 7 AUROC of liver-specific mortality and NAFLD / MASLD-related mortality obtained from the MASH prediction score, NFS, and FIB-4 among all participants in the Shanghai Changfeng cohort and the UK Biobank cohort;

[0035] Figure 8 AUROC of the MASH prediction score, NFS, and FIB-4 in predicting NAFLD / MASLD-related mortality;

[0036] Figure 9 AUROC of the MASH prediction score in predicting NAFLD / MASLD-related mortality in the subgroup analysis. Detailed implementation manners

[0037] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be further described below.

[0038] (I) Research population

[0039] All the research subjects in this embodiment are from Zhongshan Hospital, Fudan University (154 males and 157 females, with an average age of 46 years and a BMI of 27.9 kg / m 2 ) and the University of Helsinki Central Hospital in Finland (83 males and 228 females, with an average age of 49 years and a BMI of 24.3 kg / m 2 ). All the participants are adults suspected of having MASH and have undergone liver biopsy. The prediction score is further applied to 5,893 community populations in the Shanghai Changfeng Study and 111,673 in the UK Biobank to screen for MASH high-risk patients, and then the effectiveness of this non-invasive prediction model for MASH based on metabolomics in predicting the risks of all-cause death and cause-specific death in the population is evaluated.

[0040] Patients from Zhongshan Hospital Affiliated to Fudan University (China) and Helsinki University Central Hospital (Finland) were eligible to participate in this study if they met the following inclusion criteria: (1) aged between 18 and 80 years; (2) at risk of MASH, including obesity, diabetes, metabolic syndrome, or long-term elevation of serum aminotransferase concentration; (3) no other acute or chronic diseases other than obesity, type 2 diabetes, or metabolic syndrome according to medical history, physical examination, and laboratory tests. Patients with excessive alcohol intake, other liver diseases (including hepatitis B or C, autoimmune hepatitis, etc.), or those using hepatoprotective drugs such as silymarin or phosphatidylcholine were excluded.

[0041] This study was conducted under the guidance of the Declaration of Helsinki and approved by the Research Ethics Committees of Zhongshan Hospital, Fudan University and Helsinki University Central Hospital. Written informed consent was obtained from all participants.

[0042] (II) Methods and Results

[0043] 1. Liver biopsy

[0044] Liver biopsies were performed on all participants in both cohorts, and the biopsy results were interpreted by senior pathologists from both hospitals. The Steatosis, Activity, Fibrosis (SAF) score was used to evaluate the severity of non-alcoholic fatty liver disease. The SAF score consists of grades of steatosis (S, 0 - 3), activity (A, 0 - 4), and fibrosis (F, 0 - 4). The activity (A) grade is calculated by adding the grades of ballooning and lobular inflammation. If the steatosis score ≥ 1, ballooning ≥ 1, and lobular inflammation ≥ 120, MASH is diagnosed. The grading of fibrosis was performed using the Brunt criteria.

[0045] 2. Physical examination and biochemical tests

[0046] When measuring height and weight, the subjects removed their shoes, hats, and clothes, stood upright with their backs straight and feet together; the body mass index (BMI) was calculated according to the formula = weight / height (kg / m 2 ). Blood pressure was measured in the sitting position. The subjects were asked to relax and sit quietly for 5 - 10 minutes. The right upper arm of the subjects was placed on the examination table at the same height as the heart, and an electronic sphygmomanometer (OMRON Model HEM - 752FUZZY, Omron Co., Dalian, China) calibrated by a traditional mercury sphygmomanometer was used for measurement. The measurement was repeated 3 times, with an interval of at least 1 minute each time, and the average value of the 3 measurements was recorded and calculated.

[0047] Before collecting venous blood samples, the subjects were instructed to fast overnight for at least 10 hours. In the early morning of the next day, 10 ml of fasting peripheral blood was collected from the subjects' veins and placed in a serum tube. For the Chinese liver biopsy cohort and subjects in the Changfeng Research Center, complete blood count was performed using an automated hematology analyzer (ADVIA 2120, Siemens Healthcare Diagnostics, Camberley, UK); fasting plasma glucose (FPG), ALT, AST, γ-glutamyltransferase (γGT), alkaline phosphatase (ALP), total cholesterol (TC), triglyceride (TG), and high density lipoprotein cholesterol (HDL-c) were detected using an automatic biochemical analyzer (HITACHI 7600, Tokyo, Japan); LDL-c was calculated according to the Friedewald formula: LDL-c = (TC - HDL-c) - TG / 2.2; HbA 1c was detected using a high performance liquid chromatography (HPLC) (BIO-RAD II TURBO) certified by the National Glycohemoglobin Standardization Program; serum insulin content was determined using electrochemiluminescence immunoassay. The above-mentioned relevant physical examination data and biochemical indexes of the subjects in the Finnish liver biopsy cohort and the UK Biobank have been obtained previously.

[0048] 3. Based on 1 Serum metabolomics detection by 1H-NMR

[0049] Serum metabolites from the Chinese liver biopsy cohort and the Changfeng cohort were analyzed using a 600 MHz AVANCE III NMR spectrometer equipped with a BBI probe (Bruker Biospin GmbH, Germany). The detection parameters were as follows: at a temperature of 310 K, the number of data points was 98k, the number of transients was 32, the spectral width was 20 ppm, and the line broadening factor was 0.3 Hz. The one-dimensional NMR proton spectrum signals were detected using a 1D water pre-saturation NOESY pulse sequence. The NMR diffusion-edited spectrum was detected using a pre-saturated bipolar gradient longitudinal eddy current delay LEDBPGPPR2S1D pulse sequence to attenuate small molecule signals and retain macromolecule signals. After the detection was completed, all NMR spectra were pre-processed using Topspin software version 3.6.0, and the signals of all NMR spectra were integrated and quantitatively calculated based on R software version 4.3.0. The serum in the serum tube was centrifuged at 3000 rmp for 15 minutes to separate the serum, and 0.5 ml of serum was transported to the NMR detection platform at low temperature. A phosphate buffer solution (0.085 M containing 10% D2O) was prepared, and the pH was measured and adjusted to 7.38 - 7.42. The serum sample was centrifuged at 2700×g for 10 min, and 350 μL of the supernatant was mixed with 350 μL of the phosphate buffer solution. 600 μl of the mixed solution was transferred to a 5 mm NMR SampleJet tube, and the SampleJet tube was placed in the sample handling equipment. The NMR spectrometer automatically collected spectral data for each sample, and the levels of the corresponding serum metabolites were relatively quantified from the diffusion-edited spectrum. A total of 157 metabolite contents were measured in the study, and 197 metabolite-related parameters were obtained by calculation, as shown in Table 1.

[0050] Table 1 Measured metabolite contents and calculated metabolite-related parameters in the Chinese liver biopsy cohort (N = 354)

[0051]

[0052]

[0053]

[0054]

[0055]

[0056]

[0057]

[0058]

[0059] The Finnish liver biopsy cohort and the UK Biobank cohort used high-throughput NMR metabolomics analysis services (Nightingale Health Ltd., Helsinki, Finland).

[0060] The study found that about 250 metabolites and metabolism-related parameters, including lipids, lipoproteins, amino acids, ketone bodies, as well as sugar metabolites and inflammatory indicators, were simultaneously present in the liver biopsy cohorts of China and Finland.

[0061] 4. Evaluation of liver fibrosis using NFS and FIB-4

[0062] According to the management guidelines for non-alcoholic fatty liver disease jointly issued by the European Association for the Study of the Liver (EASL), the European Association for the Study of Diabetes (EASD), and the European Society for the Study of Obesity (EASO), the NFS and FIB-4 scores can effectively quantify and evaluate the degree of liver fibrosis.

[0063] NFS = -1.675 + 0.037 * age (years) + 0.094 * BMI (kg / m2) + 1.13 * impaired fasting glucose or diabetes (yes = 1, no = 0) + 0.99 * AST / ALT - 0.013 * platelet count (109 / L) - 0.66 * serum albumin (g / dL).

[0064] Those with NFS < -1.455 have a low risk of developing severe liver fibrosis; those with -0.1455 to 0.676 have a moderate risk; those with > 0.676 have a high risk.

[0065] FIB-4 = age (years) × AST (U / L) / [platelet count (109 / L) × √ALT (U / L)]. Those with FIB-4 < 1.30 have a low risk of developing severe liver fibrosis; those with 1.30 to 2.67 have a moderate risk; those with ≥ 2.67 have a high risk.

[0066] 5. Mortality and causes of death

[0067] The mortality rate of the subjects was sourced from the official data of the Shanghai Center for Disease Control and Prevention. The causes of death and the main follow-up endpoints were classified into cardiovascular diseases, liver diseases, extrahepatic tumors, and others. Liver-specific mortality was further classified into deaths caused by NAFLD / MASLD and other liver diseases. The classification criteria were based on the Tenth Revision of the International Classification of Diseases (ICD-10).

[0068] 6. Machine learning-assisted identification of candidate variables for MASH prediction

[0069] Descriptive statistics were used according to the type of variables, and univariate regression analysis was used to determine 24 clinical variables and 194 1The metabolic parameters determined by 1H-NMR were used as candidate parameters, as shown in Table 2.

[0070] Table 2 Candidate parameters related to MASH risk (N = 218)

[0071]

[0072]

[0073]

[0074]

[0075]

[0076] Note: All statistical tests were two-sided tests, and P < 0.05 was considered statistically significant.

[0077] After removing highly correlated parameters (r > 0.8) using the "findCorrelation" function in the R software package caret, a new dataset containing 91 variables was formed as candidate biomarkers for MASH. The participants in the Chinese liver biopsy cohort were randomly divided into a training set and a test set at a ratio of 7:3. The variables obtained by the random forest regression method (R software package randomForest) for the training set data were ranked according to the average Gini decrease rate. The top 15 included AST; γGT; BMI; the ratio of free cholesterol to total lipids in very low density lipoprotein (VLDL)-3 (V3FCp); the ratio of phospholipids to total lipids in total low density lipoprotein (LDL) (L0PLp); apolipoprotein E (ApoE); triglycerides in VLDL-1 (V1TG); acetic acid; triglycerides in LDL-3 (L3TG); vitamin B12; white blood cells; tryptophan; FPG; and the ratio of triglycerides to total lipids in LDL-6 (L6TGp). Further using the best subset regression method (R software package leaps), finally BMI, AST, tryptophan, and V0PLp were used as the main candidate variables for predicting MASH, with the lowest Bayesian information criterion (Bic), as Figure 1 shown. In the Chinese cohort, BMI, AST, and Tyr were significantly higher in MASH patients, while V0PLp was significantly lower, as Figure 2 shown.

[0078] 7. Development and validation of metabolome-derived MASH prediction scores

[0079] A MASH prediction model was established by multiple logistic regression using the above four candidate variables: MASH prediction score = 0.137 × BMI (kg / m2) + 0.054 × AST (U / L) + 0.033 × Tyr (μmol / L) – 0.421 × V0PLp (%) – 0.243

[0080] The area under the receiver operating characteristic curve (AUROC) was used to describe the diagnostic accuracy of the MASH prediction score in the Chinese and Finnish cohorts. The cut-off point was determined to have a sensitivity and specificity of no less than 90%.

[0081] This score was significantly correlated with hepatic steatosis, ballooning degeneration, lobular inflammation, and fibrosis scores in the Chinese cohort (P < 0.001), and the results are shown in Figure 3 ; the AUROC was 0.87 (95% CI, 0.83 - 0.91) in the Chinese cohort and 0.81 (0.75 - 0.87) in the Finnish cohort, as Figure 4 shown

[0082] The sensitivity for excluding diagnosis and the specificity for establishing diagnosis of the MASH prediction score in the Chinese cohort were both 90%, and the negative predictive value was 77.2%; applying the same cut-off value to the Finnish cohort, the sensitivity for excluding diagnosis was 96.9%, the specificity for establishing diagnosis was 79.2%, and the negative predictive value was 99.2%, as Figure 5 and Table 3 show

[0083] Table 3 Diagnostic efficacy of the metabolomics-based MASH prediction score

[0084]

[0085]

[0086] Note: All statistical tests were two-sided, and P < 0.05 was considered statistically significant

[0087] 8. The MASH prediction score is associated with liver-specific mortality

[0088] The Cox proportional hazards model was used to estimate the hazard ratios (HRs) and 95% confidence intervals (95% CIs) of all-cause and cause-specific mortality for participants at high or moderate risk of MASH (prediction score above the lower cut-off value) and those not predicted to be at risk of MASH (prediction score below the lower cut-off value) in the Shanghai Changfeng cohort and the UK Biobank cohort. Age, sex, smoking, and alcohol consumption were adjusted for in the Cox regression model

[0089] In the Shanghai Changfeng cohort, the liver-specific mortality rate in the medium- and high-risk groups, especially the mortality rate associated with NAFLD / MASLD (adjusted HR 39.41 after 5.4 years of follow-up; 95% CI 4.73 - 328.00, adjusted HR 23.19 after 7.4 years of follow-up; 95% CI 4.80 - 111.00), was significantly higher than that in the low-risk group. The mortality rates associated with cardiovascular diseases, extrahepatic tumors, and other liver diseases were similar between the two groups, as shown in Table 4. In the UK Biobank cohort, the all-cause mortality rate (adjusted HR 1.75 after 12.6 years of follow-up; 95% CI, 1.62 - 1.89) and the mortality rate associated with NAFLD / MASLD (adjusted HR 27.80 after 12.6 years of follow-up; 95% CI, 15.08 - 51.26) in the medium- and high-risk groups were also significantly higher than those in the low-risk group. The mortality rates associated with cardiovascular diseases, extrahepatic tumors, and other liver diseases were also significantly higher. The results are shown in Table 5.

[0090] Table 4 All-cause mortality and cause-specific mortality rates obtained after 5.4 years and 7.4 years of follow-up in the Shanghai Changfeng cohort with risk grouping according to the MASH prediction score (N = 5893)

[0091]

[0092]

[0093]

[0094] Table 5 All-cause mortality and cause-specific mortality rates obtained after 12.6 years of follow-up in the UK Biobank cohort with risk grouping according to the MASH prediction score (N = 111673)

[0095]

[0096]

[0097] 9. Compare the effectiveness of the MASH prediction score, NFS, and FIB-4 in predicting mortality

[0098] The DeLong algorithm was used to compare the AUROC of the MASH prediction score with that of FIB-4 and NFS. Among all participants in the Shanghai Changfeng cohort and the UK Biobank cohort, the liver-specific mortality rate (adjusted HR, 13.83; 95% CI, 10.39 - 18.41) and the mortality rate associated with NAFLD / MASLD (adjusted HR, 28.40; 95% CI, 16.09 - 50.13) in the high-risk population were significantly higher than those in the low-risk group, as Figure 6As shown, the AUROC of liver-specific mortality obtained by the MASH prediction score was 0.83 (0.83 - 0.84), and the AUROC of NAFLD / MASLD-related mortality was 0.90 (0.89 - 0.90), which was significantly superior to NFS (P = 0.01) and FIB-4 (P = 0.04) in the latter, as Figure 7 shown. And the AUROC of this score for evaluating NAFLD / MASLD-related mortality was relatively close in the two cohorts. As Figure 8 shown.

[0099] Subgroup analysis showed that after stratification by gender, age, smoking, drinking, BMI, and glucose metabolism status, this score was also applicable to the prediction of NAFLD / MASLD-related mortality (AUROCs 0.811 - 0.948), as Figure 9 shown.

[0100] (III) Summary

[0101] The present invention utilizes 24 candidate clinical parameters and 194 1 1H-NMR candidate metabolites to develop the first MASH prediction scoring model based on NMR metabolomics and machine learning techniques, and discovers that the combination of four key parameters, namely BMI, AST, tryptophan (Tyr), and the proportion of phospholipids in total lipids of VLDL (V0PLp), has good MASH prediction ability and effect. The MASH prediction scoring model was successfully validated in the Finnish liver biopsy cohort, with an AUROC of 0.81 to 0.87. This score was applied in the Finnish liver biopsy cohort and the UK Biobank cohort, and the results showed that it could identify patients with a more than 20-fold increased risk of non-alcoholic fatty liver disease / MASLD-related death, and its effect was superior to FIB-4 and NFS. These research results indicate that this score is a powerful predictor of liver-specific mortality and can be used for Chinese and Caucasian patients.

[0102] The above are only the preferred embodiments of the present invention and do not impose any limitation on the present invention. Any person skilled in the art within the technical field, without departing from the technical solution of the present invention, makes any form of equivalent replacement or modification and other changes to the technical solution and technical content disclosed by the present invention, all of which belong to the content of the technical solution of the present invention and still fall within the protection scope of the present invention.

Claims

1. A method for constructing a metabolic-related fatty hepatitis and liver disease mortality prediction model based on metabolomics, characterized in that: The prediction model uses nuclear magnetic resonance 1 The strategy of combining H-NMR metabolomics and machine learning, 1 The H-NMR serum metabolomics test was performed using a 600 MHz AVANCE Ⅲ NMR spectrometer equipped with a BBI probe, and high-throughput NMR metabolomics analysis technology was used to measure serum metabolites. 1 The detection parameters used in the H-NMR serum metabolomics test were a temperature of 310 K, a data point of 98 K, a transient of 32, a spectral width of 20 ppm, a line width factor of 0.3 Hz, The model formula is as follows: MASH prediction score = 0.137 × BMI (kg / m 2 )+0.054×AST(U / L)+0.033×Tyr(mol / L)–0.421 ×V0PLp(%)–0.

243.

2. The method for constructing a metabolic-related fatty hepatitis and liver disease death prediction model based on metabolomics according to claim 1, characterized in that: 1 H-NMR serum metabolomics detection uses a 1D water presaturation NOESY pulse sequence to detect one-dimensional NMR hydrogen spectrum signals.

3. The method for constructing a metabolic-related fatty hepatitis and liver disease death prediction model based on metabolomics according to claim 1, characterized in that: 1 H-NMR serum metabolomics detection uses a pre-saturated bipolar gradient longitudinal eddy current delay LEDBPGPPR2S1D pulse sequence to detect NMR diffusion editing spectra, which attenuates small molecule signals and retains large molecule signals.

4. The method for constructing a metabolic-related fatty hepatitis and liver disease death prediction model based on metabolomics according to claim 1, characterized in that: 1 After the H-NMR serum metabolomics test was completed, all NMR spectra were preprocessed using Topspin version 3.6.0 software, and all NMR spectral signals were integrated and quantitatively calculated based on R version 4.3.0 software.

Images