A plasma metabolic risk deviation evaluation system based on a cox proportional risk model and age calibration and a storage medium

Abstract

The application discloses a plasma metabolic risk deviation evaluation system based on a Cox proportional risk model and age calibration and a storage medium, and the system comprises the following modules: a data acquisition module, which acquires plasma metabolic marker data and time sequence age of a subject; a linear prediction score calculation module, which inputs the metabolic marker data into a pre-trained Cox proportional risk model and outputs a metabolite-driven linear prediction score; a time sequence age calibration module, which calculates an age-expected score through an age calibration function based on the time sequence age; a deviation index calculation module, which generates an age-calibrated metabolic risk deviation index according to the deviation of the metabolite-driven linear prediction score relative to the age-expected score; and an evaluation result generation module, which generates an auxiliary evaluation result based on the index. The application constructs a Cox model based on complete follow-up information, eliminates the time sequence age-related effect through age calibration, and enables the metabolic states of subjects of different ages to be compared in a unified reference framework.

Description

Technical Field

[0001] This application relates to the fields of metabolomics and bioinformatics, specifically to a plasma metabolic risk deviation assessment system and storage medium based on the Cox proportional hazards model and age calibration. Background Technology

[0002] Aging is a complex biological process in which bodily functions gradually decline over time. Metabolomics-based aging clocks analyze the metabolite profiles in biological samples such as blood and use statistical models to establish the association between metabolic patterns and time-series age or health outcomes, which can be used to assess an individual's aging status.

[0003] In existing technologies, metabolic age clocks generally use chronological age as the modeling target, employing machine learning methods to fit the relationship between plasma metabolic characteristics and chronological age, outputting metabolic age and metabolic age acceleration. Metabolic age reflects the fit between a subject's current metabolic state and chronological age, representing the relative position of metabolic state within the age distribution of the population. However, metabolic age assessment methods based on chronological age as the modeling target have limitations in clinical applications.

[0004] In the field of survival analysis, the Cox proportional hazards model has been used to calculate individual risk scores. This method uses follow-up time and follow-up event identifiers as dependent variables and metabolic biomarker data as covariates to fit a proportional hazards model, outputting an individual's metabolite-driven linear prediction score. Existing studies have reported various plasma metabolic biomarkers associated with mortality risk, covering different metabolic pathways such as amino acids, lipids, and glucose metabolites.

[0005] In the study of metabolic biomarkers, existing literature has reported the associations between phenolic microbial co-metabolites, sulfate conjugates, lipid and vitamin metabolites and the body's health status. For example, phenolic metabolites from gut microbiota have been found to be associated with cardiovascular disease and chronic kidney disease; sulfate conjugates are involved in neurotransmitter metabolism and the detoxification of exogenous substances; and certain lipid metabolites have been reported to be associated with inflammation, insulin resistance, and cellular senescence.

[0006] However, existing technologies use time-series age as the modeling target, and their core contradiction lies in the fact that the higher the model's fitting accuracy to time-series age, the more limited the incremental information that its output indicators can provide in predicting an individual's future health outcomes, which is independent of time-series age. This makes it difficult for metabolic age assessment indicators to truly reflect an individual's systemic risk level.

[0007] Therefore, there is an urgent need for a method that can reduce age-related systematic differences based on follow-up outcome-related metabolic signals and through time-series age calibration, in order to support auxiliary comparison of risk-related metabolic status among subjects of different ages. Summary of the Invention

[0008] This application provides a plasma metabolic risk deviation assessment system and storage medium based on the Cox proportional hazards model and age calibration, in order to solve the technical problem that existing metabolic age assessment methods mainly focus on time-series age fitting and are difficult to characterize risk-related metabolic differences within a peer-to-peer reference framework.

[0009] This application provides a plasma metabolic risk deviation assessment system based on the Cox proportional hazards model and age calibration, specifically including a data acquisition module, a linear prediction score calculation module, a time-series age calibration module, a deviation index calculation module, and an assessment result generation module.

[0010] The data acquisition module is used to acquire metabolic biomarker data from the subject's plasma sample and the subject's chronological age. The linear prediction score calculation module inputs the metabolic biomarker data into a pre-trained Cox proportional hazards model to calculate the subject's metabolite-driven linear prediction score. The metabolite-driven linear prediction score is a linear score output by fitting the pre-trained Cox proportional hazards model with complete follow-up time and censored information as outcomes and the metabolic biomarker data as covariates. The chronological age calibration module calculates the expected age score for the reference population at the same chronological age based on the subject's chronological age using an age calibration function. The deviation index calculation module generates an age-calibrated metabolic risk deviation index based on the deviation of the metabolite-driven linear prediction score from the expected age score. The assessment result generation module generates an auxiliary assessment result based on the age-calibrated metabolic risk deviation index using in vitro detection data.

[0011] Furthermore, the Cox proportional hazards model also includes a pre-training module for pre-training the Cox proportional hazards model, specifically including: A training dataset is obtained, comprising two or more training samples. Each training sample includes metabolic biomarker data, time-series age, follow-up time, and follow-up event identifiers. The follow-up event identifiers indicate whether a preset outcome event occurred within the corresponding follow-up time for the training sample. A Cox proportional hazards model is fitted using the metabolic biomarker data as covariates and the follow-up time and follow-up event identifiers as dependent variables to estimate a partial regression coefficient vector. The dimension of the partial regression coefficient vector is the same as the number of metabolic biomarkers in the metabolic biomarker data. L2 regularization is applied to constrain the norm of the partial regression coefficient vector. Five-fold cross-validation is used to evaluate the consistency index of the Cox proportional hazards model. After confirming the stability of the estimation of the partial regression coefficient vector based on the consistency index, the partial regression coefficient vector is stored as a parameter of the Cox proportional hazards model.

[0012] Furthermore, the formula for fitting the Cox proportional hazards model is:

[0013] in, This represents the risk function of subject X at time t. This represents the basic risk function, where n represents the number of metabolic biomarkers in the metabolic biomarker data. This represents the standardized characteristic value of the nth metabolic biomarker. This represents the partial regression coefficient of the nth metabolic biomarker.

[0014] Furthermore, the formula for calculating the metabolite-driven linear prediction score is as follows:

[0015] in, This represents the metabolite-driven linear prediction score. This represents the standardized characteristic value of the nth metabolic biomarker. This represents the partial regression coefficient of the nth metabolic biomarker.

[0016] Furthermore, the formula for calculating the age-calibrated metabolic risk deviation index is as follows:

[0017]

[0018] in, This represents the linear risk score. Indicates the age slope. Indicates chronological age. Indicates the intercept. Represents the residual. Indicates the expected score based on age. The age-calibrated metabolic risk deviation index indicates that the subject's metabolic status deviates from the expected level for their age, while an age-calibrated metabolic risk deviation index less than 0 indicates that the subject's metabolic status deviates from the expected level for their age.

[0019] Furthermore, the system also includes a data preprocessing module, which performs matrix-assisted laser desorption / ionization time-of-flight mass spectrometry (MALADS) detection on metabolic biomarker data and preprocesses the detected metabolic biomarker data. The preprocessing includes sequentially performing spectral smoothing, baseline correction, peak detection, peak alignment, batch correction, logarithmic transformation, and standardization.

[0020] Furthermore, the assessment result generation module is used to obtain the distribution position of the age-calibrated metabolic risk deviation index relative to the preset reference interval, and to map the distribution position to at least one of frailty status level, weight and metabolic status level and cardiovascular health status level, respectively, and output the auxiliary reference information of the subject.

[0021] Furthermore, the assessment result generation module is used to construct a reference function between the number of coexisting diseases and the age-calibrated metabolic risk deviation index based on training reference data; obtain the number of diagnosed chronic diseases of the subject, input the number of diagnosed chronic diseases into the reference function, and obtain the corresponding reference deviation value of coexisting diseases; when the age-calibrated metabolic risk deviation index is higher than the reference deviation value of coexisting diseases and reaches a preset threshold, output risk-related auxiliary prompts.

[0022] Furthermore, the metabolic marker data include 3-methylindole, 4-hydroxybenzoic acid, salicylic acid, 2,6-dihydroxybenzoic acid, indoleacetic acid, 3,4-dihydroxymandelic acid, vanillic acid, phenol sulfate, 4-ethylphenol sulfate, dopamine-3-O-sulfate, p-cresol sulfate, vanillic acid-4-O-sulfate, 3-hydroxybenzoic acid-3-O-sulfate, 4-methylcatechol-1-sulfate, ferulic acid-4-O-sulfate, phosphoric acid choline, sebacic acid, 2-hydroxysebacic acid, (R)-3-hydroxytetradecanoic acid, ceramide, linoleic acid, 2-hydroxystearic acid, eicosanoic acid, oleamide, vitamin A, and 11'-carboxy-γ-cryophenol.

[0023] This application also provides a computer-readable storage medium storing computer-executable instructions thereon, wherein when executed by a processor, the computer-executable instructions cause the processor to perform the following steps: The data acquisition module acquires metabolic biomarker data from the subject's plasma sample, as well as the subject's chronological age. The linear prediction score calculation module inputs the metabolic biomarker data into a pre-trained Cox proportional hazards model to calculate the subject's metabolite-driven linear prediction score. The metabolite-driven linear prediction score is a linear score output by fitting the pre-trained Cox proportional hazards model with complete follow-up time and censored information as outcomes and the metabolic biomarker data as covariates. Chronological age is not included as a covariate in the Cox proportional hazards model. The chronological age calibration module calculates the expected age score for the reference population at the same chronological age using an age calibration function based on the subject's chronological age. The deviation index calculation module generates an age-calibrated metabolic risk deviation index based on the deviation of the metabolite-driven linear prediction score from the expected age score. The evaluation result generation module generates an auxiliary evaluation result based on the age-calibrated metabolic risk deviation index using the in vitro detection data.

[0024] This application provides a plasma metabolic risk deviation assessment system and storage medium based on the Cox proportional hazards model and age calibration. By fitting the Cox proportional hazards model with complete follow-up time and censored information as the outcome and metabolic biomarker data as covariates, the system calculates the metabolite-driven linear prediction score of the subjects. It directly uses systematic risk as the prediction object, overcoming the problem of limited independent information caused by existing metabolic age clocks that use time-series age as the fitting target. Furthermore, it calculates the expected score of subjects at the same time-series age based on time-series age, generating an age-calibrated metabolic risk deviation index. This system avoids collinearity interference between time-series age and metabolic biomarkers in model fitting by excluding time-series age from the covariates of the Cox proportional hazards model, allowing the metabolite-driven linear prediction score to more purely reflect the association between metabolic biomarkers and health outcomes. Furthermore, it uses the deviation of the metabolite-driven linear prediction score from the age-predicted score as an age-calibrated metabolic risk deviation indicator, effectively reducing the systematic differences in the metabolite-driven linear prediction score related to time-series age. This enables comparisons of risk-related metabolic status among subjects of different ages within a unified reference framework. Simultaneously, it achieves a complete processing chain from plasma metabolic biomarker data input to auxiliary assessment result output, which can be used for auxiliary assessment of in vitro detection data, risk-related stratification, and health management reference. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a schematic diagram illustrating the calculation logic of the age-calibrated metabolic risk deviation index as described in the embodiments of this application; Figure 2 This is a scatter plot showing the consistency between the age-calibrated metabolic risk deviation index described in the embodiments of this application and the three clinical phenotypes; Figure 3 This is a schematic diagram illustrating the relationship between the number of coexisting diseases and the age-calibrated metabolic risk deviation index as described in the embodiments of this application; Figure 4 This is a forest plot of age-calibrated metabolic risk deviation indicators and routine blood biochemical indicators as described in the embodiments of this application; Figure 5 This is a bar chart showing the correlation between the age-calibrated metabolic risk deviation index and organ systems and specific diseases as described in the embodiments of this application; Figure 6This is a bar chart comparing the consistency index of the age-calibrated metabolic risk deviation index and the comprehensive clinical health index in predicting health outcomes, as described in the embodiments of this application. Detailed Implementation

[0027] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0028] Existing methods for assessing metabolic age mostly use time-series age as the modeling target, which can characterize the fitting relationship between plasma metabolic characteristics and age. However, their output indicators mainly reflect the deviation of metabolic age relative to time-series age. For applications that need to compare the risk-related metabolic status of subjects of different ages, relying solely on time-series age fitting models may be insufficient to fully characterize metabolic differences related to follow-up outcomes. To address this technical problem, this application includes the following technical features.

[0029] This application provides a plasma metabolic risk deviation assessment system based on time-series age calibration. Specifically, it addresses the technical problem that existing metabolic age assessment methods mainly focus on time-series age fitting and are difficult to characterize risk-related metabolic differences within a peer-to-peer reference framework. The system constructs a Cox proportional hazards model based on complete follow-up time and censored information, and uses the quantitative characteristics of predetermined metabolic biomarkers as covariates to calculate the subject's metabolite-driven linear prediction score. Based on the subject's time-series age, it calculates the expected score for the same time-series age in the reference population using an age calibration function. The deviation of the metabolite-driven linear prediction score from the expected score is used as an age-calibrated metabolic risk deviation index. Based on this index, it outputs auxiliary assessment results from in vitro detection data. The system includes the following technical features.

[0030] Specifically, the plasma metabolic risk deviation assessment system based on the Cox proportional hazards model and age calibration includes a data acquisition module, a data preprocessing module, a linear prediction score calculation module, a time-series age calibration module, a deviation index calculation module, and an assessment result generation module. Each module is described in detail below.

[0031] The data acquisition module acquires metabolic biomarker data from the subject's plasma sample, as well as the subject's chronological age.

[0032] In this application, the metabolic marker data include 3-methylindole, 4-hydroxybenzoic acid, salicylic acid, 2,6-dihydroxybenzoic acid, indoleacetic acid, 3,4-dihydroxymandelic acid, vanillic acid, phenol sulfate, 4-ethylphenol sulfate, dopamine-3-O-sulfate, p-cresol sulfate, vanillic acid-4-O-sulfate, 3-hydroxybenzoic acid-3-O-sulfate, 4-methylcatechol-1-sulfate, ferulic acid-4-O-sulfate, phosphoric acid choline, sebacic acid, 2-hydroxysebacic acid, (R)-3-hydroxytetradecanoic acid, ceramide (d18:0 / 16:0), linoleic acid, 2-hydroxystearic acid, eicosanoic acid, oleamide, vitamin A, and 11'-carboxy-γ-cryophenol.

[0033] The above 26 metabolic markers can be divided into four categories according to their metabolic functions: Phenolic co-metabolites, including 3-methylindole, 4-hydroxybenzoic acid, salicylic acid, 2,6-dihydroxybenzoic acid, indoleacetic acid, 3,4-dihydroxymandelic acid, and vanillic acid, as well as sulfated ester co-metabolites such as phenol sulfate and 4-ethylphenol sulfate, mainly originate from the metabolism of aromatic amino acids by intestinal microorganisms. Their plasma levels can reflect the co-metabolic state between intestinal microorganisms and the host.

[0034] Sulphate conjugates, including six types such as dopamine-3-O-sulfate, p-cresol sulfate, vanillic acid-4-O-sulfate, 3-hydroxybenzoic acid-3-O-sulfate, 4-methylcatechol-1-sulfate, and ferulic acid-4-O-sulfate, are involved in neurotransmitter metabolism, detoxification of exogenous substances, and secondary conversion of phenolic metabolites.

[0035] Lipid metabolites include nine types, such as phosphoric acid, sebacic acid, 2-hydroxysepiacetic acid, (R)-3-hydroxytetradecanoic acid, ceramide (d18:0 / 16:0), linoleic acid, 2-hydroxystearic acid, eicosanoic acid, and oleamide, covering fatty acid, dicarboxylic acid, ceramide, and phospholipid metabolites. Among them, ceramide has been reported to be associated with the risk of death from cardiovascular disease.

[0036] Vitamins and fat-soluble vitamin metabolites, including vitamin A (retinol) and 11'-carboxy-γ-cryophenol, participate in antioxidant defense and fat-soluble vitamin metabolic pathways.

[0037] The technical advantage lies in the fact that the combination of metabolic biomarkers focuses on metabolic pathways related to systemic risk, such as phenolic microbial cometabolites, sulfate conjugates, lipids, and vitamins. This gives the input features of the Cox proportional hazards model a clear biological orientation, rather than mixing biomarkers reflecting the current state of metabolic physiology with biomarkers reflecting systemic risk. This improves the biological interpretability of the model's input features and the accuracy of risk prediction.

[0038] Since metabolite signals in raw plasma samples are easily affected by factors such as baseline drift, random noise, retention time shifts between samples, and systematic biases between multiple batches of tests, which directly affect the accuracy and repeatability of subsequent modeling, it is necessary to standardize the data after testing.

[0039] The data preprocessing module performs matrix-assisted laser desorption / ionization time-of-flight mass spectrometry (MALADS) detection on metabolic biomarker data and preprocesses the detected metabolic biomarker data. The preprocessing includes sequentially performing spectral smoothing, baseline correction, peak detection, peak alignment, batch correction, logarithmic transformation, and standardization.

[0040] The spectrum smoothing employs a Savitzky-Golay filter to eliminate the influence of random noise on the mass spectrum peak shape.

[0041] The baseline correction employs an adaptive iterative algorithm to identify baseline points and perform polynomial fitting to subtract baseline drift from the original spectrum.

[0042] The peak detection uses continuous wavelet transform to identify the position of metabolite signal peaks, and obtains the corresponding peak position, peak height and peak area through peak shape fitting.

[0043] The peak alignment matches and aligns the mass spectrometry peaks of the same metabolite in different samples, so that the detection signals of the same metabolite in different samples can be accurately matched and compared.

[0044] The batch correction uses the ComBat method to eliminate systematic non-biological biases between multiple batches or different centers.

[0045] The logarithmic transformation is performed on the quantitative metabolite data to improve the symmetry of the data distribution.

[0046] The standardization involves subtracting the mean from the logarithmically transformed data and dividing by the standard deviation, so that the data of different metabolic markers are on the same dimension and scale.

[0047] Its technical advantage lies in eliminating non-biological noise and batch-to-batch bias in the mass spectrometry detection process, making metabolic biomarker data from different samples and batches comparable, providing standardized input features for subsequent Cox proportional hazards models, and improving the accuracy and stability of model calculations.

[0048] The linear prediction score calculation module inputs the metabolic biomarker data into a pre-trained Cox proportional hazards model to calculate the subject's metabolite-driven linear prediction score. This metabolite-driven linear prediction score is a linear score output by fitting the pre-trained Cox proportional hazards model with complete follow-up time and censoring information as outcomes and the metabolic biomarker data as covariates. The formula for calculating the metabolite-driven linear prediction score is as follows:

[0049] in, This represents the metabolite-driven linear prediction score. This represents the standardized characteristic value of the nth metabolic biomarker. This represents the partial regression coefficient of the nth metabolic biomarker.

[0050] The technical advantage lies in the fact that the metabolite-driven linear prediction score is the linear prediction component of the Cox proportional hazards model, directly reflecting the weighted contribution of each metabolic biomarker to the hazard function. This score is fitted using the complete follow-up time and censored information as the outcome, directly pointing to the healthy outcome event rather than fitting to time-series age. This overcomes the problem of limited independent information caused by existing metabolic age clocks that use time-series age as the modeling target. Simultaneously, each metabolic biomarker in the metabolite-driven linear prediction score has a clear partial regression coefficient, making the direction and intensity of each metabolic biomarker's contribution to risk quantifiable and interpretable, providing a numerical basis for subsequent age calibration processing.

[0051] In this application, the system further includes a pre-training module for pre-training the Cox proportional hazards model, specifically including: Obtain a training dataset, which contains two or more training samples. Each training sample contains metabolic biomarker data, time-series age, follow-up time, and follow-up event identifier. The follow-up event identifier is used to indicate whether a preset outcome event occurred in the training sample within the corresponding follow-up time.

[0052] Specifically, the metabolic biomarker data in the training samples contains the same types of metabolic biomarkers as the metabolic biomarker data in the data acquisition module, and is used as input features for the model; the time-series age is used to construct a subsequent age calibration function; the follow-up time is the time interval from when the subject enters the queue to when a follow-up event occurs or the last follow-up; the follow-up event identifier is used to indicate whether the training sample has experienced a preset outcome event within the corresponding follow-up time, and serves as the dependent variable for model fitting. By constructing a training dataset containing the above information, a complete data foundation including input features, survival time, and outcome information is provided for the supervised training of the Cox proportional hazards model.

[0053] Using the metabolic biomarker data as covariates and the follow-up time and follow-up event identifiers as dependent variables, a Cox proportional hazards model is fitted to estimate the partial regression coefficient vector. The dimension of the partial regression coefficient vector is the same as the number of metabolic biomarkers in the metabolic biomarker data. The formula for fitting the Cox proportional hazards model is:

[0054] in, This represents the risk function of subject X at time t. This represents the basic risk function, where n represents the number of metabolic biomarkers in the metabolic biomarker data. This represents the standardized characteristic value of the nth metabolic biomarker. This represents the partial regression coefficient of the nth metabolic biomarker.

[0055] Specifically, the basic risk function It does not rely on the subject's metabolic biomarker data and reflects the risk trend over time at baseline levels; the index part This represents the relative contribution of each metabolic biomarker to the risk function, where the partial regression coefficient is... A positive value indicates that an increase in the standardized characteristic value of this metabolic marker indicates an increased risk. A negative value indicates that the risk decreases when the standardized characteristic value of the metabolic marker increases.

[0056] In this application, the time-series age is not included as a covariate in the Cox proportional hazards model. Unlike existing techniques that include time-series age and metabolic biomarkers as covariates in the model, this application excludes time-series age from model fitting. This ensures that the partial regression coefficients estimated by the model only reflect the association between each metabolic biomarker and the outcome event, avoiding collinearity interference between time-series age and metabolic biomarkers in model fitting. By maximizing the partial likelihood function to estimate the partial regression coefficient vector, subjects who experienced the event at the time point of observation of the follow-up event have a higher risk probability than subjects still in the risk cluster. At this point, the metabolite-driven linear prediction score output by the model is entirely driven by metabolic biomarker data and does not include the model-estimated component of time-series age, thus preserving independent and complete age-related information for subsequent age calibration.

[0057] Since metabolic biomarker data contains multiple metabolic biomarkers, there may be collinearity among them, and the model input features have high dimensionality. Direct model fitting can easily lead to overfitting. Therefore, it is necessary to introduce L2 regularization to constrain the norm of the partial regression coefficient vector to reduce the risk of overfitting.

[0058] The norm of the partial regression coefficient vector is constrained by L2 regularization.

[0059] Specifically, an L2 penalty term (penalizer=0.1) is added to the objective function of the Cox proportional hazards model. This penalty term constrains the Cox proportional hazards model to maximize the partial likelihood function while also considering the norm of the partial regression coefficient vector, thus shrinking the estimated partial regression coefficient vector towards zero. Through L2 regularization, the model's oversensitivity to noise and fluctuations in the measurement of specific metabolic biomarkers in the training data is reduced, the risk of overfitting is decreased, and the predictive stability of the model on new samples is improved.

[0060] After constraining the norm of the partial regression coefficient vector through L2 regularization, it is necessary to evaluate the generalization ability of the partial regression coefficient vector on new samples to confirm the stability of its estimation. Therefore, this application uses five-fold cross-validation to evaluate the fitted Cox proportional hazards model.

[0061] Five-fold cross-validation is used to evaluate the consistency index of the Cox proportional hazards model. After confirming the stability of the partial regression coefficient vector based on the consistency index, the partial regression coefficient vector is stored as the parameter of the Cox proportional hazards model. This allows the linear prediction score calculation module to be directly called without repeated training, ensuring the reproducibility of the evaluation results and computational efficiency.

[0062] Specifically, the training dataset is randomly divided into 5 non-overlapping equal subsets. One subset is selected as the validation set and the other 4 subsets are selected as the training set. This process is repeated 5 times. In each round, the Cox proportional hazards model is refitted on the training set and the consistency index is calculated on the corresponding validation set. Finally, 5 consistency indices are obtained, and the average value is taken as the comprehensive evaluation index of the model performance.

[0063] The consistency index (C-index) is a commonly used evaluation metric for measuring the predictive ability of the Cox proportional hazards model. It is used to assess the degree of consistency between the individual risk order predicted by the model and the actual observed survival time order. The consistency index ranges from 0.5 to 1, where a value of 0.5 indicates that the model has no predictive ability (random guessing), a value of 1 indicates that the model is completely consistent (the prediction perfectly matches the reality), and a higher value indicates better predictive performance. In this application, when the average consistency index of the five-fold cross-validation is higher than a preset threshold, the estimation of the partial regression coefficient vector is confirmed to be stable.

[0064] The advantage of the pre-training process is that it forms a complete technical chain from data construction, parameter estimation, regularization constraints, stability verification to parameter solidification, providing a validated and reusable model parameter foundation for calculating the metabolite-driven linear prediction score of any subject.

[0065] The temporal age calibration module calculates the expected age score of the reference population at the same temporal age based on the subject's temporal age using an age calibration function.

[0066] Because the metabolite-driven linear prediction score may include a component related to time series age, even subjects of similar metabolic status may exhibit systematic differences in their metabolite-driven linear prediction scores due to age, resulting in a lack of cross-comparison of scores between individuals of different age groups. To eliminate this age-related effect, this application uses age calibration to calibrate the time series age-related component of the metabolite-driven linear prediction score. Specifically, based on the time series age and metabolite-driven linear prediction score of each training sample in the training set, a simple linear regression equation is fitted to establish a reference relationship between the metabolite-driven linear prediction score and the time series age, which serves as the age calibration function. The subject's time series age is then substituted into the age calibration function to calculate the expected age score for the same time series age in the reference population. The expression for the age calibration function is:

[0067] in, The linear risk score, representing the metabolite-driven linear prediction score in the age-calibrated function, is numerically equal to the metabolite-driven linear prediction score. Both represent different expressions of the same value at different treatment stages. It is emphasized that this value comes from the predictive output of the Cox proportional hazards model. It is emphasized that this value appears as a dependent variable in the age calibration function.

[0068] The deviation index calculation module generates an age-calibrated metabolic risk deviation index (Age Deviation) based on the deviation of the metabolite-driven linear prediction score from the age-expected score.

[0069] The formula for calculating the age-calibrated metabolic risk deviation index is as follows:

[0070] in This represents the age slope, i.e., the slope of the age calibration function. Indicates chronological age. Indicates the intercept. This indicates the age-calibrated residual. This represents the expected score for the same age group in the reference population. The age-calibrated metabolic risk deviation index indicates that the subject's metabolite-driven linear prediction score is higher than the expected level in the reference population at the same age, while the age-calibrated metabolic risk deviation index is less than 0, indicating that the subject's metabolite-driven linear prediction score is lower than the expected level in the reference population at the same age.

[0071] In this application, the correlation coefficient between the age-calibrated metabolic risk deviation index and time-series age approaches 0, effectively eliminating the correlation effect of time-series age and making the age-calibrated metabolic risk deviation index of individuals in different age groups horizontally comparable. Simultaneously, the age-calibrated metabolic risk deviation index is judged with 0 as a benchmark, transforming the abstract metabolite-driven linear prediction score into a relative deviation measure referenced to the average level of the same age group, providing a directly interpretable quantitative indicator for the assessment result generation module.

[0072] like Figure 1 The diagram illustrates the overall process from inputting metabolic biomarker data, calculating metabolite-driven linear prediction scores using the Cox proportional hazards model, and then outputting an age-calibrated metabolic risk deviation index after age calibration. Figure 1 Figure A illustrates the overall process from inputting metabolic biomarker data to outputting an age-calibrated metabolic risk deviation index. The left side shows the data on 56 metabolites and all-cause mortality outcomes obtained from the follow-up population; the middle section calculates the mortality risk score using a Cox proportional hazards model; the scatter plot below uses time-series age as the horizontal axis and mortality risk score as the vertical axis, with the red line representing the age-expected score (fitted based on the reference population), and the arrow indicating "Age deviation = Risk score - Age expectation," which is the age-calibrated metabolic risk deviation index; the right side shows the correlation between this index and multi-system disease burden and clinical indicators.

[0073] Figure 1 Figure B shows the distribution of the logarithmic mortality risk with age bias. The horizontal axis represents age bias (i.e., the age-calibrated metabolic risk deviation index), and the vertical axis represents the logarithmic mortality risk. The red horizontal line represents the reference benchmark with an age bias of 0, and the red arrows indicate the age bias of a subject, representing the degree of deviation of their logarithmic mortality risk from the expected level for their age group.

[0074] Figure 1 C: This shows the density distribution of age bias in the population. The horizontal axis represents age bias, and the vertical axis represents density. The distribution is approximately normal, with the center located near 0, indicating that the age bias of most individuals is close to 0, while a few individuals have significant positive or negative biases.

[0075] The assessment result generation module generates auxiliary assessment results based on the age-calibrated metabolic risk deviation index, using in vitro detection data.

[0076] In this application, the assessment result generation module is used to obtain the distribution position of the age-calibrated metabolic risk deviation index relative to a preset reference interval, and to map the distribution position to at least one of frailty status level, weight metabolic status level and cardiovascular health status level, and output auxiliary reference information of the subject.

[0077] In this application, the preset reference interval is predetermined based on the age-calibrated metabolic risk deviation index distribution of the training set samples.

[0078] In this application, the frailty level is a classification result of a comprehensive frailty index constructed based on the principle of cumulative health defects, used to reflect the degree of frailty of the subject; the weight and metabolic status level is a classification result based on body mass index (BMI), used to reflect the subject's body composition metabolic status; and the cardiovascular health status level is a classification result based on the LE8 cardiovascular health score proposed by the American Heart Association, used to reflect the subject's cardiovascular health status.

[0079] By mapping the distribution of the age-calibrated metabolic risk deviation index relative to the preset reference interval to the aforementioned state levels, a quantitative association between the age-calibrated metabolic risk deviation index and multiple recognized clinical phenotypes is achieved. This demonstrates that the index integrates metabolic health information from multiple organ systems, possesses excellent comprehensive assessment capabilities, and transforms the assessment results from an abstract numerical form into a readable report based on clinically recognized indicators, making it easier for clinicians and subjects to understand.

[0080] In this application, the age-calibrated metabolic risk deviation index is significantly positively correlated with the frailty index (regression coefficient β>0, P<0.001), indicating that a higher age-calibrated metabolic risk deviation index suggests a higher degree of frailty; the age-calibrated metabolic risk deviation index is significantly positively correlated with body mass index (regression coefficient β>0, P<0.001), indicating that being overweight is positively associated with the accumulation of metabolic risk; the age-calibrated metabolic risk deviation index is significantly negatively correlated with the LE8 cardiovascular health score (regression coefficient β<0, P<0.001), indicating that a better cardiovascular health status corresponds to a lower age-calibrated metabolic risk deviation index; the absolute values ​​of all three correlation coefficients are not less than 0.30.

[0081] like Figure 2 As shown, this is a scatter plot of the consistency between the metabolite coefficients, an age-calibrated metabolic risk deviation indicator, and the metabolite coefficients of the three clinical phenotypes, where: Figure 2Figure A shows the consistency between the metabolite coefficients (horizontal axis, β (age bias)) of the age-calibrated metabolic risk deviation index and the metabolite coefficients (vertical axis, β (weakness index)) of the frailty index. The scatter plots show a positive correlation trend, the red regression line has a positive slope, and the consistency correlation coefficient r = 0.500 (P = 8.85 × 10⁻⁶). -5 The figure shows representative metabolites such as 3-methylindole, 4-hydroxybenzoic acid, 2-hydroxysepiacaloic acid, phosphocholine, and salicylic acid.

[0082] Figure 2 Figure B shows the consistency between the metabolite coefficients of the age-calibrated metabolic risk deviation index (horizontal axis, β (age deviation)) and the metabolite coefficients of the LE8 cardiovascular health score (vertical axis, β (LE8)). The scatter plots show a negative correlation trend, the slope of the red regression line is negative, and the consistency correlation coefficient r = 0.392 (P = 0.003). Representative metabolites such as 1,6-dihydroxybenzoic acid, sebacic acid, indoleacetic acid, oleic acid amide, and salicylic acid are marked in the figure.

[0083] Figure 2 The middle section (C) shows the consistency between the metabolite coefficients of the age-calibrated metabolic risk deviation index (horizontal axis, β(age deviation)) and the metabolite coefficients of body mass index (BMI) (vertical axis, β(BMI)). The scatter plots show a positive correlation trend, the slope of the red regression line is positive, and the consistency correlation coefficient r = -0.325 (P = 0.014). Representative metabolites such as 4-hydroxybenzoic acid, (R)-3-hydroxytetradecanoic acid, 2-hydroxystearic acid, ceramide Cer(d18:0 / 16:0), and eicosanoic acid are marked in the figure.

[0084] In this application, the assessment result generation module is used to construct a reference function between the number of coexisting diseases and the age-calibrated metabolic risk deviation index based on training reference data; obtain the number of diagnosed chronic diseases of the subject, input the number of diagnosed chronic diseases into the reference function, and obtain the corresponding reference deviation value of coexisting diseases; when the age-calibrated metabolic risk deviation index is higher than the reference deviation value of coexisting diseases and reaches a preset threshold, output risk-related auxiliary prompts.

[0085] In this application, the preset threshold is predetermined based on the standard deviation of the age-calibrated metabolic risk deviation index in the training set samples.

[0086] Specifically, the number of diagnosed chronic disease types and their corresponding age-calibrated metabolic risk deviation indices are obtained for each training sample in the training set. Using the number of diagnosed chronic disease types as the independent variable and the age-calibrated metabolic risk deviation index as the dependent variable, a linear regression model is fitted to estimate the regression coefficients (the increment of the age-calibrated metabolic risk deviation index for each additional chronic disease). This linear regression model is used as the reference function. The reference function reflects the average level of the age-calibrated metabolic risk deviation index under different chronic disease burdens in the training reference data; that is, the age-calibrated metabolic risk deviation value expected solely from the number of coexisting diseases, without additional metabolic risk deviation.

[0087] Based on this, the number of diagnosed chronic diseases of the subject is obtained, and the number of diagnosed chronic diseases is input into the reference function to calculate the reference deviation value for coexistence of multiple diseases for the subject under the corresponding number of chronic diseases. This value represents the expected age-calibrated metabolic risk deviation level of the subject under the same chronic disease burden, and serves as a benchmark for judging whether its metabolic risk deviation exceeds the explanatory range of the coexistence burden of multiple diseases. When the subject's age-calibrated metabolic risk deviation index is higher than the reference deviation value for coexistence of multiple diseases by a preset threshold (e.g., one standard deviation), it indicates that the subject's actual metabolic risk deviation level is significantly higher than the average level of the population with the same chronic disease burden, suggesting that its metabolic status has an additional deviation beyond the explanation of the coexistence burden of multiple diseases. At this time, a risk-related auxiliary prompt is output.

[0088] like Figure 3 As shown, based on a disease cohort of n=7,768, β=0.0366 (95% CI: 0.0273–0.0459, P=1.6×10⁻⁶). -14 The linear relationship is robustly established within the range of 0-7 diseases, demonstrating a linear early warning benchmark with an increase of 0.0366 in the age-calibrated metabolic risk deviation index for each additional chronic disease.

[0089] In addition to the coexistence of multiple diseases, this application further verified the association between age-calibrated metabolic risk deviation indicators and routine clinical blood biochemical indicators, organ system disease burden, and mortality prediction ability.

[0090] like Figure 4As shown, after adjusting for age and sex per SD: creatinine β=0.1390 ​​(95% CI: 0.1253-0.1527, P<0.001), blood urea nitrogen β=0.1126 (95% CI: 0.1010-0.1242, P<0.001), and uric acid β=0.0810 (P<0.001) were significantly positive; hemoglobin β=-0.0736 (P<0.001), hematocrit β=-0.0679 (P<0.001), and red blood cell count β=-0.0532 (P<0.001) were significantly negative, demonstrating a significant positive association between age-calibrated metabolic risk deviation indicators and indicators such as creatinine, blood urea nitrogen, and uric acid, and a significant negative association with indicators such as hemoglobin and hematocrit.

[0091] like Figure 5 As shown, the β values ​​for the endocrine system were 0.0641 (95% CI: 0.0406-0.0875, P<0.001), the renal system β value was 0.0571 (95% CI: 0.0048-0.1093, P<0.05), the cardiovascular system β value was 0.0405 (95% CI: 0.0114-0.0696, P<0.01), and the nervous system β value was 0.0373 (95% CI: 0.0021-0.0726, P<0.05). Specific diseases included: chronic kidney disease β value was 0.1924 (95% CI: 0.0920-0.2928, P<0.001), and hyperuricemia / gout β value was 0.1130 (95% CI: 0.0920-0.2928, P<0.001). CI: 0.0755-0.1505, P<0.001), β=0.0743 for stroke (95% CI: 0.0338-0.1148, P<0.001), demonstrating the positive association between the burden of endocrine, renal, cardiovascular, and nervous system diseases and age-calibrated metabolic risk deviation indicators, as well as the effect strength of specific diseases such as chronic kidney disease, hyperuricemia / gout, and stroke.

[0092] like Figure 6 The figure shows the comparison results of age-calibrated metabolic risk deviation index and comprehensive clinical health index in predicting health outcomes, where: Figure 6Table A shows a comparison of the hazard ratios (HRs) of the age-calibrated metabolic risk deviation index, the frailty index, and the LE8 cardiovascular health score. The "Individual Models" section above shows the results of each index individually included in the Cox model: HR for the age-calibrated metabolic risk deviation index = 1.53 (per standard deviation, P < 0.001), HR for the frailty index = 1.26 (P < 0.001), and HR for the LE8 cardiovascular health score = 0.89 (P < 0.05). The "Mutual Correction" section below shows the results after all three are included in the model simultaneously: HR for the age-calibrated metabolic risk deviation index = 1.50 (P < 0.001), HR for the frailty index = 1.22 (P < 0.001), and HR for the LE8 cardiovascular health score = 0.93 (no statistical significance), indicating that the age-calibrated metabolic risk deviation index maintains its independent predictive ability after correcting for the other composite indicators.

[0093] Figure 6 Figure B shows a comparison of the C-index values ​​for each model. The univariate C-index for age-calibrated metabolic risk deviation was 0.7343, frailty index was 0.7226, and LE8 cardiovascular health score was 0.7105. After adding the age-calibrated metabolic risk deviation to the "frailty index + LE8" baseline model, the C-index increased from 0.7235 to 0.7453 (LRT P = 7.3 × 10⁻⁶). -22 This indicates that the age-calibrated metabolic risk deviation index has significant incremental predictive value for existing comprehensive clinical indicator models.

[0094] Figure 6 C: This section shows the univariate hazard ratio comparison between the age-calibrated metabolic risk deviation index and various routine blood biochemical indicators. The HR of the age-calibrated metabolic risk deviation index (blue) was 1.53 (P<0.001), which was significantly better than or independent of biochemical indicators such as ALT, UREA, CREA, UA, HDLC, LDLC, TC, TG, and GLU, while some of these biochemical indicators did not show statistical significance.

[0095] Figure 6 Figure D: This demonstrates the independent predictive power of age-calibrated metabolic risk deviation indicators after adjusting for each routine blood biochemical indicator. The red dots represent the HR (risk advance) of each biochemical indicator itself, and the blue squares represent the HR of the age-calibrated metabolic risk deviation indicator after simultaneously incorporating that biochemical indicator. The results show that after adjusting for ALT, UREA, CREA, UA, HDLC, LDLC, TC, TG, and GLU, the age-calibrated metabolic risk deviation indicators all maintained a significant positive association (P < 0.001), indicating that they carry predictive information independent of routine blood biochemical indicators.

[0096] Figure 6Figure E shows the changes in the C-index after incorporating the age-calibrated metabolic risk deviation index into various biochemical indicator models. Light pink bars represent the C-index of models using only the corresponding biochemical indicator, purple dots represent the C-index of models combining the age-calibrated metabolic risk deviation index and the corresponding biochemical indicator, and dashed lines represent the C-index of the age-calibrated metabolic risk deviation index as a single factor. The C-index of each combined model was significantly higher than that of models using only biochemical indicators (P<0.001), further demonstrating that the age-calibrated metabolic risk deviation index has stable incremental predictive value for routine clinical indicators.

[0097] Based on the same technical concept as the above system embodiments, the technical solution of this application can also be implemented in the form of a computer-readable storage medium to adapt to deployment and operation on different computing devices. The storage medium stores computer-executable instructions, which, when executed by a processor, can fully realize the aforementioned core functions such as data acquisition, linear prediction score calculation, age calibration, deviation index calculation, and evaluation result generation. This storage medium can be packaged as an independent software installation package and deployed on a local device, or it can be integrated into an existing clinical testing information system to provide standardized calculation tools for the auxiliary evaluation of in vitro testing data.

[0098] Specifically, this application also provides a computer-readable storage medium storing computer-executable instructions thereon, wherein when executed by a processor, the computer-executable instructions cause the processor to perform the following steps: Step S1: The data acquisition module acquires metabolic marker data from the subject's plasma sample and the subject's chronological age.

[0099] In step S2, the linear prediction score calculation module inputs the metabolic biomarker data into a pre-trained Cox proportional hazards model to calculate the subject's metabolite-driven linear prediction score. The metabolite-driven linear prediction score is a linear score output by fitting the pre-trained Cox proportional hazards model with complete follow-up time and censored information as the outcome and the metabolic biomarker data as the covariate; wherein, the time-series age is not included as a covariate in the Cox proportional hazards model.

[0100] Step S3: The time-series age calibration module calculates the expected age score of the reference population at the same time-series age based on the subject's time-series age using an age calibration function.

[0101] Step S4: The deviation index calculation module generates an age-calibrated metabolic risk deviation index based on the deviation of the metabolite-driven linear prediction score relative to the age-expected score.

[0102] Step S5: The evaluation result generation module generates auxiliary evaluation results based on the age-calibrated metabolic risk deviation index using in vitro detection data.

[0103] Furthermore, in the storage medium, the processor also performs the following steps: Step S6: The pre-training module pre-trains the Cox proportional hazards model. Specifically, the pre-training process includes: Step S61: Obtain a training dataset. The training dataset contains two or more training samples. Each training sample contains metabolic biomarker data, time-series age, follow-up time, and follow-up event identifier. The follow-up event identifier is used to indicate whether a preset outcome event occurs in the training sample within the corresponding follow-up time.

[0104] Step S62: Using the metabolic biomarker data as covariates and the follow-up time and follow-up event identifier as dependent variables, fit a Cox proportional hazards model to estimate a partial regression coefficient vector. The dimension of the partial regression coefficient vector is the same as the number of metabolic biomarkers in the metabolic biomarker data.

[0105] Step S63: Apply L2 regularization to constrain the norm of the partial regression coefficient vector.

[0106] Step S64: Use five-fold cross-validation to evaluate the consistency index of the Cox proportional hazards model. After confirming that the estimation of the partial regression coefficient vector is stable based on the consistency index, store the partial regression coefficient vector as the parameters of the Cox proportional hazards model.

[0107] Furthermore, in the storage medium, the processor also performs the following steps: Step S7: The data preprocessing module performs matrix-assisted laser desorption / ionization time-of-flight mass spectrometry (MALADS) detection on the metabolic biomarker data and preprocesses the detected metabolic biomarker data. The preprocessing includes sequentially performing spectral smoothing, baseline correction, peak detection, peak alignment, batch correction, logarithmic transformation, and standardization.

[0108] The advantage of this application is that it provides a plasma metabolic risk deviation assessment system and storage medium based on the Cox proportional hazards model and age calibration. By fitting the Cox proportional hazards model with complete follow-up time and censored information as the outcome and metabolic biomarker data as covariates, it calculates the metabolite-driven linear prediction score of the subjects and directly uses systematic risk as the prediction object, overcoming the problem of limited independent information caused by the existing metabolic age clock which uses time-series age as the fitting target; and it calculates the expected score of subjects at the same time-series age based on time-series age to generate an age-calibrated metabolic risk deviation index. This system avoids collinearity interference between time-series age and metabolic biomarkers in model fitting by excluding time-series age from the covariates of the Cox proportional hazards model, allowing the metabolite-driven linear prediction score to more purely reflect the association between metabolic biomarkers and health outcomes. Furthermore, it uses the deviation of the metabolite-driven linear prediction score from the age-predicted score as an age-calibrated metabolic risk deviation indicator, effectively reducing the systematic differences in the metabolite-driven linear prediction score related to time-series age. This enables comparisons of risk-related metabolic status among subjects of different ages within a unified reference framework. Simultaneously, it achieves a complete processing chain from plasma metabolic biomarker data input to auxiliary assessment result output, which can be used for auxiliary assessment of in vitro detection data, risk-related stratification, and health management reference.

[0109] The foregoing provides a detailed description of a plasma metabolic risk deviation assessment system and storage medium based on the Cox proportional hazards model and age calibration. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A plasma metabolic risk deviation assessment system based on the Cox proportional hazards model and age calibration, characterized in that, Specifically, it includes: The data acquisition module acquires metabolic marker data from the subject's plasma sample, as well as the subject's chronological age. The linear prediction score calculation module inputs the metabolic biomarker data into a pre-trained Cox proportional hazards model to calculate the subject's metabolite-driven linear prediction score. The metabolite-driven linear prediction score is a linear score output by fitting the pre-trained Cox proportional hazards model with complete follow-up time and censoring information as outcomes and the metabolic biomarker data as covariates; wherein, the temporal age is not included as a covariate in the Cox proportional hazards model. The temporal age calibration module calculates the expected age score of the reference population at the same temporal age based on the subject's temporal age using an age calibration function. The deviation index calculation module generates an age-calibrated metabolic risk deviation index based on the deviation of the metabolite-driven linear prediction score from the age-expected score; and The assessment result generation module generates auxiliary assessment results based on the age-calibrated metabolic risk deviation index, using in vitro detection data.

2. The plasma metabolic risk deviation assessment system based on the Cox proportional hazards model and age calibration as described in claim 1, characterized in that, It also includes a pre-training module for pre-training the Cox proportional hazards model, specifically including: Obtain a training dataset, which contains two or more training samples. Each training sample contains metabolic biomarker data, time-series age, follow-up time, and follow-up event identifier. The follow-up event identifier is used to indicate whether a preset outcome event occurred in the training sample within the corresponding follow-up time. Using the metabolic biomarker data as covariates and the follow-up time and follow-up event identifier as dependent variables, a Cox proportional hazards model is fitted to estimate a partial regression coefficient vector. The dimension of the partial regression coefficient vector is the same as the number of metabolic biomarkers in the metabolic biomarker data. The norm of the partial regression coefficient vector is constrained by L2 regularization; The consistency index of the Cox proportional hazards model is evaluated using five-fold cross-validation. After confirming the stability of the estimate of the partial regression coefficient vector based on the consistency index, the partial regression coefficient vector is stored as the parameter of the Cox proportional hazards model.

3. The plasma metabolic risk deviation assessment system based on the Cox proportional hazards model and age calibration as described in claim 2, characterized in that, The formula for fitting the Cox proportional hazards model is: ； in, This represents the risk function of subject X at time t. This represents the basic risk function, where n represents the number of metabolic biomarkers in the metabolic biomarker data. This represents the standardized characteristic value of the nth metabolic biomarker. This represents the partial regression coefficient of the nth metabolic biomarker.

4. The plasma metabolic risk deviation assessment system based on the Cox proportional hazards model and age calibration as described in claim 1, characterized in that, The formula for calculating the metabolite-driven linear prediction score is as follows: ； in, This represents the metabolite-driven linear prediction score. This represents the standardized characteristic value of the nth metabolic biomarker. This represents the partial regression coefficient of the nth metabolic biomarker.

5. The plasma metabolic risk deviation assessment system based on the Cox proportional hazards model and age calibration as described in claim 1, characterized in that, The time-series age is used to construct the age calibration function, and the calculation formula for the age-calibrated metabolic risk deviation index is as follows: ； ； in, This represents the linear risk score. Indicates the age slope. Indicates chronological age. Indicates the intercept. Represents the residual. Indicates the expected score based on age. The age-calibrated metabolic risk deviation index indicates that the subject's metabolic status deviates from the expected level for their age, while an age-calibrated metabolic risk deviation index less than 0 indicates that the subject's metabolic status deviates from the expected level for their age.

6. The plasma metabolic risk deviation assessment system based on the Cox proportional hazards model and age calibration as described in claim 1, characterized in that, Also includes The data preprocessing module performs matrix-assisted laser desorption / ionization time-of-flight mass spectrometry (MALADS) detection on metabolic biomarker data and preprocesses the detected metabolic biomarker data. The preprocessing includes sequentially performing spectral smoothing, baseline correction, peak detection, peak alignment, batch correction, logarithmic transformation, and standardization.

7. The plasma metabolic risk deviation assessment system based on the Cox proportional hazards model and age calibration as described in claim 1, characterized in that, The evaluation result generation module is used for, The distribution position of the age-calibrated metabolic risk deviation index relative to a preset reference interval is obtained, and the distribution position is mapped to at least one of frailty status level, weight and metabolic status level and cardiovascular health status level, respectively, and the auxiliary reference information of the subject is output.

8. The plasma metabolic risk deviation assessment system based on the Cox proportional hazards model and age calibration as described in claim 1, characterized in that, The evaluation result generation module is used for, A reference function was constructed based on the training reference data to establish the relationship between the number of coexisting diseases and the age-calibrated metabolic risk deviation index. The number of diagnosed chronic diseases of the subject is obtained, and the number of diagnosed chronic diseases is input into the reference function to obtain the corresponding reference deviation value for coexistence of multiple diseases; When the age-calibrated metabolic risk deviation index exceeds the multi-disease coexistence reference deviation value and reaches a preset threshold, a risk-related auxiliary prompt is output.

9. The plasma metabolic risk deviation assessment system based on the Cox proportional hazards model and age calibration as described in claim 1, characterized in that, The metabolic marker data include 3-methylindole, 4-hydroxybenzoic acid, salicylic acid, 2,6-dihydroxybenzoic acid, indoleacetic acid, 3,4-dihydroxymandelic acid, vanillic acid, phenol sulfate, 4-ethylphenol sulfate, dopamine-3-O-sulfate, p-cresol sulfate, vanillic acid-4-O-sulfate, 3-hydroxybenzoic acid-3-O-sulfate, 4-methylcatechol-1-sulfate, ferulic acid-4-O-sulfate, phosphocholine, sebacic acid, 2-hydroxysebacic acid, (R)-3-hydroxytetradecanoic acid, ceramide, linoleic acid, 2-hydroxystearic acid, eicosanoic acid, oleamide, vitamin A, and 11'-carboxy-γ-cryophenol.

10. A computer-readable storage medium having stored thereon computer-executable instructions, wherein, When this computer-executable instruction is executed by the processor, it causes the processor to perform the following steps: The data acquisition module acquires metabolic marker data from the subject's plasma sample, as well as the subject's chronological age; The linear prediction score calculation module inputs the metabolic biomarker data into a pre-trained Cox proportional hazards model to calculate the subject's metabolite-driven linear prediction score. The metabolite-driven linear prediction score is a linear score output by fitting the pre-trained Cox proportional hazards model with complete follow-up time and censoring information as outcomes and the metabolic biomarker data as covariates; wherein, the time-series age is not included as a covariate in the Cox proportional hazards model. The temporal age calibration module calculates the expected age score of the reference population at the same temporal age based on the subject's temporal age using an age calibration function; The deviation index calculation module generates an age-calibrated metabolic risk deviation index based on the deviation of the metabolite-driven linear prediction score relative to the age-expected score; and The assessment result generation module generates auxiliary assessment results based on the age-calibrated metabolic risk deviation index, using in vitro test data.

Images