Disease risk prediction method and device based on plasma proteomics and waist circumference

Abstract

This invention discloses a method and device for predicting disease risk based on plasma proteomics and waist circumference, belonging to the field of medical data processing. Utilizing 2,920 plasma proteins from 52,879 participants in the UK Biobank, this invention constructs a proteomics-based waist circumference proWC score using LASSO regression, and derives the protein-anthropometric difference index proWCΔ to quantify the difference between proteomics-based waist circumference and measured waist circumference. Based on the proWC score and proWCΔ, disease risk association analysis, mortality prediction, causal mediation analysis, and dietary association analysis were conducted across 14 organ systems, demonstrating that proWC has superior clinical predictive value compared to waist circumference alone. Furthermore, based on proWC, a hidden high-risk subgroup was identified—patients with normal measured waist circumference but a near-overtly obese risk of diabetes-related death. By constructing proWCΔ and proWC, this invention elevates the assessment of central obesity from morphological measurements to the molecular level, providing a new molecular framework for precise risk stratification of central obesity.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, and in particular to a method and apparatus for predicting disease risk based on plasma proteomics and waist circumference. Background Technology

[0002] Waist circumference (WC) is a core anthropometric measure for clinical assessment of central obesity and is closely associated with the risk of type 2 diabetes, cardiovascular disease, and all-cause mortality. However, as a morphological measurement, waist circumference only reflects the physical dimension of the abdomen and cannot capture the complex biological heterogeneity behind the same waist circumference. This limitation is particularly prominent in two metabolic phenotypes: one is metabolically healthy obesity (MHO), i.e., individuals with significantly elevated waist circumference but relatively normal metabolic indicators; the other is metabolically unhealthy normal weight (MUNW), i.e., individuals with normal waist circumference but significant insulin resistance, dyslipidemia, and cardiovascular risk. The risk of diabetes in MUNW individuals is comparable to that of overt obesity, yet they are systematically missed under current anthropometric screening standards.

[0003] In recent years, the rapid development of high-throughput proteomics technology has provided new opportunities to elucidate the molecular heterogeneity of central obesity. Circulating proteomics integrates genetic information with real-time environmental exposure and can reflect the endocrine and paracrine functional status of adipose tissue. However, the proteomic architecture of central obesity has not yet been systematically elucidated, and methods for constructing waist circumference alternative scores based on proteomics for precise risk stratification are currently lacking.

[0004] Currently, existing methods have the following main drawbacks:

[0005] The assessment of central obesity is limited to a single dimension: traditional waist circumference measurement only provides static morphological information and cannot reflect the metabolic activity and inflammatory state of adipose tissue. It cannot identify metabolic obesity and normal weight phenotypes, resulting in a significant proportion of high-risk individuals being missed.

[0006] Lack of alternative indicators for waist circumference based on molecular markers: Current proteomics research focuses on the discovery of biomarkers for specific diseases, and there is no framework to integrate proteomics information systems into equivalent molecular readings of waist circumference, thus limiting the application of proteomics data in clinical risk stratification.

[0007] The difference between proteomics and morphological waist circumference cannot be quantified: Existing methods cannot quantify the difference between an individual's measured waist circumference and their proteomics-predicted waist circumference, which itself carries important pathophysiological information about the degree of metabolic imbalance.

[0008] Unidentified high-risk subgroups: Current clinical screening relies on waist circumference thresholds (e.g., ≥90 cm for men and ≥80 cm for women), which cannot identify individuals with normal waist circumference but high metabolic burden as indicated by proteomics, resulting in a significant risk of missed diagnosis. Summary of the Invention

[0009] To address the above problems, this invention proposes a method and device for predicting disease risk based on plasma proteomics and waist circumference. By training a LASSO regression prediction model using proteomics data and applying it to waist circumference prediction, two indicators are constructed: waist circumference residual (proWCΔ) and proteomics-based waist circumference (proWC). This elevates the assessment of central obesity from morphological measurements to a molecular level. It enables the use of proteomics data for clinical risk stratification; reflects the metabolic activity and inflammatory state of adipose tissue; and quantifies the difference between an individual's measured waist circumference and their proteomics-predicted waist circumference, carrying important pathophysiological information about the degree of metabolic imbalance.

[0010] On the one hand, the disease risk prediction method based on plasma proteomics and waist circumference involves the following steps:

[0011] S1, Obtain a dataset including proteomics data, waist circumference (WC) measurements, age, sex, and outcome;

[0012] S2, using the proteomics data in the dataset as the independent variable, age and gender as covariates, and waist circumference measurement as the dependent variable, train the LASSO regression prediction model to obtain the trained LASSO regression prediction model.

[0013] S3. Input the dataset into the trained LASSO regression prediction model to obtain the proteomics-predicted waist circumference pWC; use the proteomics-predicted waist circumference pWC as the dependent variable and the waist circumference measurement value WC as the independent variable to perform linear regression fitting and obtain the fitting equation.

[0014] S4, substitute the waist circumference measurement value WC into the fitting equation to obtain the expected waist circumference; calculate the difference between the proteomics predicted waist circumference pWC and the expected waist circumference to obtain the waist circumference residual proWCΔ; add the waist circumference residual proWCΔ to the waist circumference measurement value to obtain the proteomics waist circumference proWC.

[0015] S5, Disease risk stratification based on proteomics waist circumference proWC and outcomes in the dataset;

[0016] S6. Input the proteomics data, waist circumference measurement (WC), age, and gender of the individual to be evaluated into the trained LASSO regression prediction model to obtain the proteomics-predicted waist circumference (pWC) of the individual to be evaluated; and calculate the waist circumference residual (proWCΔ) and proteomics-based waist circumference (proWC) of the individual to be evaluated based on the fitted equation.

[0017] S7. Based on the proteomics waist circumference (proWC) of the individual to be evaluated, determine the corresponding risk value in the disease risk stratification to complete the risk assessment.

[0018] Preferably, the optimization objective of the LASSO regression prediction model is expressed as:

[0019] ;

[0020] in, This represents an estimate of the protein coefficient vector; denoted by ; y represents the measured waist circumference vector; X represents the standardized protein expression matrix; Z represents the age and sex covariate matrix; β represents the protein coefficient vector; γ represents the covariate coefficient vector; λ represents the regularization parameter selected through 10-fold cross-validation to minimize the mean squared error; argmin represents the parameter value that minimizes the function. Represents the L1 norm; This represents the L2 norm.

[0021] Preferably, the fitting equation is expressed as:

[0022] ;

[0023] Where pWC represents proteomics-predicted waist circumference; Indicates the intercept; ε represents the slope; WC represents the waist circumference measurement; ε represents the error term.

[0024] Preferably, the step of stratifying disease risk based on proteomics waist circumference (proWC) and outcomes in the dataset specifically involves:

[0025] Sort all proteomics waist circumference proWC values ​​in the dataset from low to high; then divide them into several quantiles.

[0026] For the binary outcome of whether the disease under study occurred during the pre-defined observation period, the disease outcome was used as the dependent variable, proteomics waist circumference (proWC) as the independent variable, and age and gender as covariates. A multivariate logistic regression model was used to output the odds ratio, 95% confidence interval, and p-value of each quantile relative to the lowest risk quantile group. If there is a quantile with a p-value less than the pre-defined threshold and an odds ratio 95% confidence interval that does not include 1, then proteomics waist circumference (proWC) is considered to be "significantly associated" with the disease under study; otherwise, the association is considered "insignificant."

[0027] Preferably, it also includes: S8, performing disease association analysis on WC, proWCΔ, and proWC; specifically: for the disease to be studied, performing disease risk stratification on waist circumference residual proWCΔ and waist circumference measurement value WC respectively; wherein, the covariate of waist circumference residual proWCΔ also includes waist circumference measurement value; establishing a disease set for each disease to be studied that is "significantly associated" with waist circumference residual proWCΔ, waist circumference measurement value WC, and proteomics waist circumference proWC; mapping the three disease sets to a three-dimensional Venn diagram for quantitative analysis.

[0028] Preferably, the quantiles are quintiles.

[0029] Preferably, step S7 further includes: performing joint stratified analysis and identifying occult high-risk subgroups; specifically:

[0030] Using the median WC and median proWC as classification thresholds, the dataset was divided into four groups: Group 1: low WC + low proWC; Group 2: low WC + high proWC; Group 3: high WC + low proWC; Group 4: high WC + high proWC. The four groups were then used as a whole in a Cox proportional hazards model, with age and gender as covariates. The hazard ratios and 95% confidence intervals of Groups 2, 3, and 4 relative to Group 1 were output. Group 2 is a high-risk subgroup for concealed infection.

[0031] Preferably, it also includes: S8, using WC, proWC, and proWCΔ to perform dietary association analysis; specifically: using dietary intake as the exposure variable, with whole age, sex, race, education level, Townsend deprivation index, smoking, and alcohol consumption as covariates, and WC, proWC, or proWCΔ as the outcome variable, performing multivariate linear regression; using dietary intake as the exposure variable, with whole age, sex, race, education level, Townsend deprivation index, smoking, alcohol consumption, and BMI as covariates, and WC, proWC, or proWCΔ as the outcome variable, performing multivariate linear regression; the diet that shows a significant association between the outcome variable and diet in both multivariate linear regressions is taken as a proteomics-specific dietary signal.

[0032] Preferably, it also includes: S8, using proWC to perform protein causal mediation analysis; the specific steps are as follows:

[0033] In Phase 1, least squares regression was performed with candidate proteins as independent variables, BMI as covariate, and proWC as dependent variable to screen out proteins with significance levels lower than the preset significance threshold.

[0034] In Phase Two, Cox proportional hazards regression was performed with candidate proteins as independent variables, disease outcomes as dependent variables, and BMI as covariate; proteins with significance levels below the preset significance threshold were screened out.

[0035] In Phase 3, a counterfactual framing causal mediation analysis was performed on the common proteins screened in Phases 1 and 2 to identify proteins with statistically significant mediating effects.

[0036] On the other hand, a disease risk prediction device based on plasma proteomics and waist circumference includes the following:

[0037] The dataset acquisition module is used to acquire datasets including proteomics data, waist circumference (WC) measurements, age, sex, and outcomes.

[0038] The regression prediction model training module is used to train the LASSO regression prediction model with proteomics data in the dataset as independent variables, age and gender as covariates, and waist circumference measurement as dependent variable, and obtain the trained LASSO regression prediction model.

[0039] The fitting equation acquisition module is used to input the dataset into the trained LASSO regression prediction model to obtain the proteomics-predicted waist circumference pWC; with the proteomics-predicted waist circumference pWC as the dependent variable and the waist circumference measurement value WC as the independent variable, a linear regression fitting is performed to obtain the fitting equation.

[0040] The proteomics waist circumference acquisition module is used to substitute the waist circumference measurement value WC into the fitting equation to obtain the expected waist circumference; the difference between the proteomics predicted waist circumference pWC and the expected waist circumference is used to obtain the waist circumference residual proWCΔ; the waist circumference residual proWCΔ is added to the waist circumference measurement value to obtain the proteomics waist circumference proWC.

[0041] The disease risk stratification module is used to stratify disease risk based on the proteomics waist circumference proWC and outcomes of the dataset;

[0042] The proteomics waist circumference acquisition module for the individual to be evaluated is used to obtain the proteomics waist circumference proWC of the individual to be evaluated by the proteomics data, waist circumference measurement (WC), age and sex of the individual to be evaluated through the fitting equation acquisition module and the proteomics waist circumference proWC acquisition module.

[0043] The individual risk assessment module is used to determine the corresponding risk value in the disease risk stratification based on the individual's proteomics waist circumference (proWC) and complete the risk assessment.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] (1) The proWC scoring system proposed in this invention integrates the systematic information of 2,920 plasma proteins, which elevates the assessment of central obesity from morphological measurement to molecular reading level. It has been validated in external independent cohorts (Geographic Validation Cohort in Scotland / Wales, UK; GNHS Mass Spectrometry Proteomics Cohort in China) and has robust generalization performance across platforms and populations.

[0046] (2) The proWCΔ index constructed in this invention can systematically quantify the difference between proteomics and morphological waist circumference, identify a hidden high-risk subgroup with normal measured waist circumference but high proteomics risk, whose diabetes mortality risk is close to that of patients with overt obesity, and break through the limitations of the current anthropometric screening standards.

[0047] (3) This invention elevates proWC from a composite risk score to a molecular tool with mechanistic insights through disease-specific causal mediation analysis, identifying disease-specific protein mediators such as FABP4, ADM, PRAP1, IGSF9 and IGFBP2, providing a hypothetical framework for targeted intervention.

[0048] (4) The present invention uses proWC for diet association analysis. The results show that proWC can reflect molecular metabolic changes related to diet patterns in addition to traditional anthropometric indicators, thereby capturing the molecular metabolic imprint of diet patterns on adipose tissue independent of weight changes, and has the potential to serve as a biomarker for early response to nutritional intervention. Attached Figure Description

[0049] The present invention will now be described in further detail with reference to the accompanying drawings;

[0050] Figure 1 This is a flowchart of a disease risk prediction method based on plasma proteomics and waist circumference according to an embodiment of the present invention.

[0051] Figure 2 This is a flowchart illustrating the design and analysis of a disease risk prediction method based on plasma proteomics and waist circumference, as described in an embodiment of the present invention.

[0052] Figure 3 The image shows a scatter plot of waist circumference (blue for males, orange for females) for a disease risk prediction method based on plasma proteomics and waist circumference according to an embodiment of the present invention. Specifically, a represents a scatter plot of LASSO predicted waist circumference (pWC) versus measured waist circumference (WC) in the complete cohort; b represents a scatter plot of proWC versus WC in the complete cohort; c represents a scatter plot of pWC versus WC in the independent Scottish and Welsh cohorts; and d represents a scatter plot of proWC versus WC in the independent Scottish and Welsh cohorts.

[0053] Figure 4Forest plot of the four joint groups and the risk ratios of all-cause mortality, cardiovascular mortality, cancer mortality and diabetes mortality for the disease risk prediction method based on plasma proteomics and waist circumference in this embodiment of the invention.

[0054] Figure 5 This is a structural block diagram of a disease risk prediction device based on plasma proteomics and waist circumference according to an embodiment of the present invention. Detailed Implementation

[0055] The present invention will be further described below through specific embodiments.

[0056] like Figure 1 and Figure 2 As shown, the disease risk prediction method based on plasma proteomics and waist circumference involves the following steps:

[0057] S1, Obtain a dataset including proteomics data, waist circumference (WC) measurements, age, sex, and outcome;

[0058] S2, using the proteomics data in the dataset as the independent variable, age and gender as covariates, and waist circumference measurement as the dependent variable, train the LASSO regression prediction model to obtain the trained LASSO regression prediction model.

[0059] S3. Input the dataset into the trained LASSO regression prediction model to obtain the proteomics-predicted waist circumference pWC; use the proteomics-predicted waist circumference pWC as the dependent variable and the waist circumference measurement value WC as the independent variable to perform linear regression fitting and obtain the fitting equation.

[0060] S4, substitute the waist circumference measurement value WC into the fitting equation to obtain the expected waist circumference; calculate the difference between the proteomics predicted waist circumference pWC and the expected waist circumference to obtain the waist circumference residual proWCΔ; add the waist circumference residual proWCΔ to the waist circumference measurement value to obtain the proteomics waist circumference proWC.

[0061] S5, Disease risk stratification based on proteomics waist circumference proWC and outcomes in the dataset;

[0062] S6. Input the proteomics data, waist circumference measurement (WC), age, and gender of the individual to be evaluated into the trained LASSO regression prediction model to obtain the proteomics-predicted waist circumference (pWC) of the individual to be evaluated; and calculate the waist circumference residual (proWCΔ) and proteomics-based waist circumference (proWC) of the individual to be evaluated based on the fitted equation.

[0063] S7. Determine the corresponding risk value in the disease risk stratification based on the individual's proteomics waist circumference to complete the risk assessment.

[0064] Specifically, the dataset used in this embodiment is from the UK Biobank (UKB), a prospective cohort study that enrolled over 500,000 participants aged 40–69 from the UK between 2006 and 2010. This embodiment includes 52,879 participants with complete proteomics data and waist circumference measurements, divided geographically into a discovery cohort (England, n=46,994) and an internal validation cohort (Scotland + Wales, n=5,885). Additionally, the Guangzhou Nutrition and Health Study (GNHS) cohort (n=485, based on mass spectrometry proteomics) in China was used as an independent external validation cohort across platforms. Exclusion criteria: Participants missing waist circumference, proteomics data, or necessary covariates (age, sex).

[0065] Proteomics data from the UK Biobank were analyzed using the Olink Explore 3072 platform (proximal extension assay, PEA technology) to detect 2,923 plasma proteins, with protein expression levels as... Standardized relative unit reporting. After quality control, 2,920 proteins that passed quality control were retained for model construction. The GNHS cohort proteomics data were obtained using liquid chromatography-mass spectrometry (LC-MS / MS). Its protein detection panels differ from and have limited overlap with the UKB platform, and were used to evaluate the cross-platform generalization ability of the proWC framework.

[0066] Specifically, the construction process of the LASSO regression prediction model is as follows:

[0067] Using the standardized expression levels of 2,920 plasma proteins as penalized predictive variables, age and sex as non-penalized covariates, and measured waist circumference as the dependent variable, the following optimization objective was constructed:

[0068] ;

[0069] in, This represents the measured waist circumference vector. This represents the normalized protein expression matrix (P=2,920). Represents the age and gender covariate matrix. This is a vector of protein coefficients. λ is the covariate coefficient vector, and λ is the regularization parameter selected through 10-fold cross-validation to minimize the mean squared error; each feature variable is standardized before model training.

[0070] Specifically, the calculation steps for proWC in this embodiment are as follows:

[0071] Proteomics-based waist circumference prediction. The expression levels of 2,920 standardized plasma proteins, along with age and sex, were input into a LASSO regression model. Ten-fold cross-validation was used to select the regularization parameter λ that minimized the mean squared error, yielding the proteomics-predicted waist circumference (pWC). Model performance was evaluated using R² and root mean square error (RMSE).

[0072] proWCΔ derivation. Fitting a linear regression of pWC to the measured WC on the entire dataset ( ), calculate the individual residuals as proWCΔ (i.e., protein-anthropometric difference).

[0073] proWC calculation. Adding proWCΔ back to the measured WC, we get proWC = WC + proWCΔ, which makes the slope of proWC close to 1.00, providing a molecular reading anchored to the measured waist circumference. Positive proWCΔ indicates that the proteomics-predicted waist circumference is higher than the measured waist circumference (metabolic obesity phenotype), while negative proWCΔ indicates that the proteomics-predicted waist circumference is lower than the measured waist circumference (metabolic relative protection phenotype).

[0074] Biological characterization. The effect sizes of WC and proWC were compared using whole-proteome association analysis (normalized β coefficient, BH-FDR correction). The per-protein effect size slope (proWC / WC) was calculated to verify the systematic amplification effect of proWC on biological signals (slope = 1.115). Z-score gradients of six key proteins (LEP, SSC4D, IGFBP1, OXT, GH1, FGF21) were analyzed using proWC Δ quantiles. See also... Figure 3 As shown. Figure 3 In the figure, 'a' represents a scatter plot of LASSO-predicted waist circumference (pWC) versus measured waist circumference in the complete cohort (n = 52,879), including the regression line, equation, and R² value; dots are colored by biological sex (blue: male, orange: female). A systematic sex-dependent difference exists, with male participants tending to cluster above the identity line (pWC = WC) (pWC > WC), while female participants cluster below the identity line (pWC < WC). This pattern is termed Proteomics-Anthropometry Sex Bimorphism (PASD) in this study. Marginal histograms show the sex-stratified frequency distribution of waist circumference and pWC. Figure 3 In the middle, b represents the scatter plot of proWC versus measured waist circumference in the complete discovery cohort, showing higher predictive accuracy (R² = 0.858) and a fit slope close to 1; the PASD pattern still exists, with male participants (blue) clustering above the isoline and female participants (orange) clustering below the isoline. The edge histogram shows the gender stratification distribution. Figure 3In the middle, c and d represent external validation of pWC and proWC in independent Scottish and Welsh cohorts (n = 5,885), confirming that the PASD pattern has geographical universality, i.e., in both indicators, male participants (blue) consistently cluster above the identity line, while female participants (orange) cluster below the identity line.

[0075] Specifically, this embodiment uses the following method to perform disease risk stratification: Participants are divided into five quantiles (Q1–Q5) and 95% confidence intervals according to WC, proWC, or proWCΔ. Using Q1 as a reference, multivariate logistic regression is used to estimate the OR value (odds ratio) for diseases with ≥100 events. The WC and proWC models are adjusted for age and gender, and the proWCΔ model is further adjusted for measured waist circumference. The P value is corrected for the Benjamini-Hochberg false discovery rate, and the significance threshold is defined as corrected P<0.05 and confidence interval not containing 1.

[0076] Specifically, this embodiment also uses WC, proWC, and proWCΔ to predict mortality and identify hidden high-risk subgroups. Using all-cause mortality, cardiovascular disease (CVD) mortality, cancer mortality, and diabetes-related mortality as outcomes, a restricted cubic spline Cox model (4 degrees of freedom) was used to assess the continuous variable mortality associations of WC, proWC, and proWCΔ, with the sample mean as a reference. All models were adjusted for age and sex, and the proWCΔ model was additionally adjusted for measured waist circumference. ANOVA was used to test for nonlinearity; Schoenfeld residuals were used to verify the proportional hazard hypothesis; a forest plot was drawn using a quantile model (Q1 reference) to quantify the HR and 95% CI for each quantile. The medians of WC and proWC were used as cutoff thresholds to construct a four-group joint stratification (low WC + high proWC; high WC + low proWC; high WC + high proWC; low WC + low proWC [reference]).

[0077] Identification of occult high-risk subgroups was conducted. Using median waist circumference (WC) and proWC as classification thresholds, a low WC + high proWC group (measured waist circumference below median but proWC above median) was identified as an occult high-risk subgroup (low WC + high proWC, accounting for 6.6%, n=3,474). Metabolic phenotypic characteristics (HbA1c, fasting blood glucose, triglycerides, HDL-cholesterol) of this subgroup were assessed and compared with a high WC + high proWC reference group. The Cox model was used to quantify the all-cause, cardiovascular, cancer, and diabetes mortality risks of this subgroup. Risk of morbidity in 14 ICD-10 disease categories was also assessed. See the Manhattan chart for (risk ratio). Figure 4 As shown, solid symbols indicate significance after BH correction.

[0078] Specifically, this embodiment also conducts protein causal mediation analysis using WC, proWC, and proWCΔ. Counterfactual framework causal mediation analysis (CMA) was performed on 30 candidate proteins and 12 major disease outcomes. The average causal mediation effect (ACME), average direct effect (ADE), and mediation proportion (PM = ACME / total effect) were estimated using the quasi-Bayesian Monte Carlo approximation method (1,000 simulations).

[0079] A three-stage mediation analysis process was used to systematically evaluate the mediating role of 30 candidate proteins in the proWC-disease association pathway under a BMI-adjusted model: The first stage screened proteins significantly associated with proWC (OLS regression, Bonferroni corrected P < 0.05); the second stage screened proteins significantly associated with disease outcome (Cox regression, Bonferroni corrected P < 0.05); the third stage performed counterfactual causal mediation analysis (quasi-Bayesian Monte Carlo approximation, 1,000 simulations) on proteins that met the criteria of both stages, estimating ACME, ADE, and PM (=ACME / total effect). A PM exceeding 100% indicated an inhibitory effect (direct and indirect effects are in opposite directions).

[0080] Specifically, this embodiment also uses WC, proWC, and proWCΔ to characterize PASD and analyze its association with diet. PASD is characterized and its association with diet is analyzed through the following steps:

[0081] (1) PASD characterization: violin plots of proWCΔ(proWC-WC) were plotted stratified by age (37–72 years, 5-year groups), and box plots of WC and proWC were plotted according to BMI categories (underweight, normal weight, overweight, obese). The interaction effects of sex × age and sex × BMI were systematically evaluated. The Level effect (baseline difference) and Shape effect (age trajectory difference) were tested using a linear mixed-effects model, and both were considered significant with P < 0.001. The reproducibility of PASD was validated in an internal validation cohort in Scotland / Wales and an external cohort in GNHS (FDR < 0.05 was considered significant).

[0082] (2) Dietary association analysis: Using the baseline intake of 15 dietary factors from the UK Biobank as exposure variables and WC, proWC, and proWCΔ as outcomes, multivariate linear regression was performed (Model 1: adjusted for age, sex, race, education level, Townsend deprivation index, smoking, and alcohol consumption; Model 2: additionally adjusted for BMI based on Model 1). The criterion for determining proteomics-specific dietary signals was that the association remained significant after BMI adjustment (BH-corrected P < 0.05). Sensitivity analysis (significant interaction test) was performed on beef, oily fish, and processed meat products stratified by sex, age, smoking status, and BMI.

[0083] The model performance was validated.

[0084] The proWC scoring system proposed in this invention achieved the following main performance results: In the discovery cohort, pWC explained 79.1% of the measured WC (R²=0.791); after residual anchoring based on the measured WC, proWC showed higher consistency with the measured WC (R²=0.858), and the fitting slope was close to 1.00. In the Scotland / Wales validation cohort, pWC had an R²=0.793, and proWC had an R²=0.858; in the GNHS independent cross-platform validation cohort, although pWC was limited by protein panel coverage (R²=0.145), proWC achieved an adjusted R² of 0.896, and the per-protein effect size slope was 1.232 (R²=0.925), demonstrating that the proWC residual anchoring procedure remains highly robust under limited protein panel conditions. The disease associations identified by proWC in the highest risk quantile (Q5) were proWC-specific in approximately 66% of cases, and it demonstrated superior predictive power to WC across all four mortality outcomes.

[0085] In summary, the proWC scoring system proposed in this invention integrates systematic information from 2,920 plasma proteins, elevating the assessment of central obesity from morphological measurements to molecular readings. It has been validated in external independent cohorts (UK Scotland / Wales geographic validation cohort n=5,885, R²=0.858; China GNHS mass spectrometry proteomics cohort n=485, R²=0.896), demonstrating robust generalization across platforms and populations.

[0086] The proWCΔ index constructed in this embodiment of the invention can systematically quantify the differences between proteomics and morphological waist circumference, and identify a hidden high-risk subgroup that accounts for 6.6% of the population (n=3,474), with normal measured waist circumference but high proteomics risk. Their risk of death from diabetes (HR≈3.5) is close to that of patients with overt obesity, which breaks through the limitations of the current anthropometry screening standards.

[0087] This invention elevates proWC from a composite risk score to a molecular tool with mechanistic insights through disease-specific causal mediation analysis, identifying disease-specific protein mediators such as FABP4 (30.5% mediation rate for I25; 117.6% mediation rate for N18), ADM (36.6% mediation rate for I25), PRAP1, IGSF9, and IGFBP2, providing a hypothetical framework for targeted intervention.

[0088] This invention discloses a novel biological phenomenon—proteomics-anthropometric sex bimorphism (PASD): males systematically have higher proWC than measured WC (suggesting that waist circumference underestimates male molecular metabolic risk), while females have the opposite (suggesting that waist circumference overestimates female molecular metabolic risk); this pattern has been validated in all age groups and two independent external cohorts, providing a new molecular basis for sex-specific metabolic risk assessment;

[0089] In this embodiment of the invention, proWC was used for dietary association analysis. The results showed that after adjusting for BMI, proWC still retained a positive association with red meat (especially beef and lamb) and a protective association with plant-based foods, while the corresponding WC association was significantly weakened. This indicates that proWC can capture the molecular metabolic imprint of dietary patterns on adipose tissue independent of weight changes and has the potential to serve as an early response biomarker for nutritional interventions.

[0090] like Figure 5 As shown, the present invention also discloses a disease risk prediction device based on plasma proteomics and waist circumference, comprising:

[0091] The dataset acquisition module 501 is used to acquire datasets including proteomics data, waist circumference measurements (WC), age, sex, and outcomes.

[0092] The regression prediction model training module 502 is used to train the LASSO regression prediction model with proteomics data in the dataset as independent variables, age and gender as covariates, and waist circumference measurement as dependent variable, so as to obtain the trained LASSO regression prediction model.

[0093] The fitting equation acquisition module 503 is used to input the dataset into the trained LASSO regression prediction model to obtain the proteomics-predicted waist circumference pWC; with the proteomics-predicted waist circumference pWC as the dependent variable and the waist circumference measurement value WC as the independent variable, a linear regression fitting is performed to obtain the fitting equation.

[0094] The proteomics waist circumference acquisition module 504 is used to substitute the waist circumference measurement value WC into the fitting equation to obtain the expected waist circumference; to obtain the waist circumference residual proWCΔ by subtracting the proteomics predicted waist circumference pWC from the expected waist circumference; and to obtain the proteomics waist circumference proWC by adding the waist circumference residual proWCΔ to the waist circumference measurement value.

[0095] Disease risk stratification module 505 is used to stratify disease risk based on the proteomics waist circumference proWC and outcomes of the dataset;

[0096] The proteomics waist circumference acquisition module 506 for the individual to be evaluated is used to input the proteomics data, waist circumference measurement value WC, age and gender of the individual to be evaluated into the trained LASSO regression prediction model to obtain the proteomics predicted waist circumference pWC of the individual to be evaluated; and to calculate the waist circumference residual proWCΔ and proteomics waist circumference proWC of the individual to be evaluated based on the fitting equation.

[0097] The individual risk assessment module 507 is used to determine the corresponding risk value in the disease risk stratification based on the individual's proteomics waist circumference, and to complete the risk assessment.

[0098] The specific implementation of the disease risk prediction device based on plasma proteomics and waist circumference is the same as that of the disease risk prediction method based on plasma proteomics and waist circumference, and will not be described again in this embodiment.

[0099] The above are merely specific embodiments of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial modifications made to the present invention using this concept shall be considered as infringing upon the protection scope of the present invention.

Claims

1. A method for predicting disease risk based on plasma proteomics and waist circumference, characterized in that, Includes the following steps: S1, Obtain a dataset including proteomics data, waist circumference (WC) measurements, age, sex, and outcome; S2, using the proteomics data in the dataset as the independent variable, age and gender as covariates, and waist circumference measurement as the dependent variable, train the LASSO regression prediction model to obtain the trained LASSO regression prediction model. S3. Input the dataset into the trained LASSO regression prediction model to obtain the proteomics-predicted waist circumference pWC; use the proteomics-predicted waist circumference pWC as the dependent variable and the waist circumference measurement value WC as the independent variable to perform linear regression fitting and obtain the fitting equation. S4, substitute the waist circumference measurement value WC into the fitting equation to obtain the expected waist circumference; calculate the difference between the proteomics predicted waist circumference pWC and the expected waist circumference to obtain the waist circumference residual proWCΔ; add the waist circumference residual proWCΔ to the waist circumference measurement value to obtain the proteomics waist circumference proWC. S5, Disease risk stratification based on proteomics waist circumference proWC and outcomes in the dataset; S6. Input the proteomics data, waist circumference measurement (WC), age, and gender of the individual to be evaluated into the trained LASSO regression prediction model to obtain the proteomics-predicted waist circumference (pWC) of the individual to be evaluated; and calculate the waist circumference residual (proWCΔ) and proteomics-based waist circumference (proWC) of the individual to be evaluated based on the fitted equation. S7. Determine the corresponding risk value in the disease risk stratification based on the individual's proteomics waist circumference to complete the risk assessment.

2. The disease risk prediction method based on plasma proteomics and waist circumference according to claim 1, characterized in that, The optimization objective of the LASSO regression prediction model is expressed as: ; in, This represents an estimate of the protein coefficient vector; denoted by ; y represents the measured waist circumference vector; X represents the standardized protein expression matrix; Z represents the age and sex covariate matrix; β represents the protein coefficient vector; γ represents the covariate coefficient vector; λ represents the regularization parameter selected through 10-fold cross-validation to minimize the mean squared error; argmin represents the parameter value that minimizes the function. Represents the L1 norm; This represents the L2 norm.

3. The disease risk prediction method based on plasma proteomics and waist circumference according to claim 1, characterized in that, The fitting equation is expressed as: ； Where pWC represents proteomics-predicted waist circumference; Indicates the intercept; ε represents the slope; WC represents the waist circumference measurement; ε represents the error term.

4. The disease risk prediction method based on plasma proteomics and waist circumference according to claim 1, characterized in that, The disease risk stratification based on proteomics waist circumference (proWC) and outcomes in the dataset is as follows: Sort all proteomics waist circumference proWC values ​​in the dataset from low to high; then divide them into several quantiles. For the binary outcome of whether the disease under study occurred during the pre-defined observation period, the disease outcome was used as the dependent variable, and proteomics waist circumference (proWC) was used as the independent variable. Age and gender were used as covariates. A multivariate logistic regression model was adopted to output the odds ratio of each quantile relative to the lowest risk quantile group, the 95% confidence interval, and the p-value. If there is a p-value less than the preset threshold and a quantile where the 95% confidence interval of the ratio does not contain 1, then the proteomics waist circumference proWC is "significantly associated" with the disease under study; otherwise, the association is "not significant".

5. The disease risk prediction method based on plasma proteomics and waist circumference according to claim 4, characterized in that, Also includes: S8. Disease association analysis was performed on WC, proWCΔ, and proWC. Specifically, for the diseases under study, disease risk stratification was performed on the waist circumference residual proWCΔ and waist circumference measurement WC, respectively. The covariates of waist circumference residual proWCΔ also included waist circumference measurement. A disease set was established for each disease under study that showed a significant association with waist circumference residual proWCΔ, waist circumference measurement WC, and proteomics-based waist circumference proWC. The three disease sets were then mapped onto a three-dimensional Venn diagram for quantitative analysis.

6. The disease risk prediction method based on plasma proteomics and waist circumference according to claim 4, characterized in that, The aforementioned quantiles are quintiles.

7. The disease risk prediction method based on plasma proteomics and waist circumference according to claim 1, characterized in that, S7 further includes: performing joint stratified analysis and identifying occult high-risk subgroups; specifically: Using the median WC and median proWC as classification thresholds, the dataset was divided into four groups: Group 1: low WC + low proWC; Group 2: low WC + high proWC; Group 3: high WC + low proWC; Group 4: high WC + high proWC. The four groups were then used as a whole in a Cox proportional hazards model, with age and gender as covariates. The hazard ratios and 95% confidence intervals of Groups 2, 3, and 4 relative to Group 1 were output. Group 2 is a high-risk subgroup for occult disease.

8. The disease risk prediction method based on plasma proteomics and waist circumference according to claim 1, characterized in that, Also includes: S8. Dietary association analysis was performed using WC, proWC, and proWCΔ. Specifically: Multivariate linear regression was performed with dietary intake as the exposure variable, whole age, sex, race, education level, Townsend deprivation index, smoking, and alcohol consumption as covariates, and WC, proWC, or proWCΔ as the outcome variable. Multivariate linear regression was also performed with dietary intake as the exposure variable, whole age, sex, race, education level, Townsend deprivation index, smoking, alcohol consumption, and BMI as covariates, and WC, proWC, or proWCΔ as the outcome variable. Diets that showed a significant association between the outcome variable and diet in both multivariate linear regressions were considered as proteomics-specific dietary signals.

9. The disease risk prediction method based on plasma proteomics and waist circumference according to claim 1, characterized in that, Also includes: S8. Use proWC for protein causal mediation analysis; the specific steps are as follows: In Phase 1, least squares regression was performed with candidate proteins as independent variables, BMI as covariate, and proWC as dependent variable to screen out proteins with significance levels lower than the preset significance threshold. In Phase Two, Cox proportional hazards regression was performed with candidate proteins as independent variables, disease outcomes as dependent variables, and BMI as covariate; proteins with significance levels below the preset significance threshold were screened out. In Phase 3, a counterfactual framing causal mediation analysis was performed on the common proteins screened in Phases 1 and 2 to identify proteins with statistically significant mediating effects.

10. A disease risk prediction device based on plasma proteomics and waist circumference, characterized in that, Including the following: The dataset acquisition module is used to acquire datasets including proteomics data, waist circumference (WC) measurements, age, sex, and outcomes. The regression prediction model training module is used to train the LASSO regression prediction model with proteomics data in the dataset as independent variables, age and gender as covariates, and waist circumference measurement as dependent variable, and obtain the trained LASSO regression prediction model. The fitting equation acquisition module is used to input the dataset into the trained LASSO regression prediction model to obtain the proteomics-predicted waist circumference pWC; with the proteomics-predicted waist circumference pWC as the dependent variable and the waist circumference measurement value WC as the independent variable, a linear regression fitting is performed to obtain the fitting equation. The proteomics waist circumference acquisition module is used to substitute the waist circumference measurement value WC into the fitting equation to obtain the expected waist circumference; the difference between the proteomics predicted waist circumference pWC and the expected waist circumference is used to obtain the waist circumference residual proWCΔ; the waist circumference residual proWCΔ is added to the waist circumference measurement value to obtain the proteomics waist circumference proWC. The disease risk stratification module is used to stratify disease risk based on the proteomics waist circumference proWC and outcomes of the dataset; The proteomics waist circumference acquisition module for the individual to be evaluated is used to input the proteomics data, waist circumference measurement (WC), age, and gender of the individual to be evaluated into a trained LASSO regression prediction model to obtain the proteomics predicted waist circumference (pWC) of the individual to be evaluated; and to calculate the waist circumference residual (proWCΔ) and proteomics waist circumference (proWC) of the individual to be evaluated based on the fitted equation. The individual risk assessment module is used to determine the corresponding risk value in the disease risk stratification based on the individual's proteomics waist circumference, and to complete the risk assessment.

Images