Establishment method and application of late-onset psoriasis risk prediction model

Abstract

The invention relates to the technical field of disease risk prediction and precision medical treatment, in particular to an establishment method and application of a late-onset psoriasis risk prediction model, and the method comprises the following steps: (1) collecting lifestyle data, serum metabolite data, clinical characteristics and polygene risk scores of a subject; (2) constructing a healthy lifestyle score according to the lifestyle data; (3) screening metabolites significantly related to the healthy lifestyle by using a multiple linear regression model; (4) screening metabolites significantly related to the risk of late psoriasis by using a Cox regression model; (5) screening a key metabolite set by adopting an elastic network regression model; and (6) inputting the sample features into a machine learning model to obtain a late-onset psoriasis risk prediction model. According to the method, the prediction accuracy is high, the AUC can reach 0.86 by combining the lifestyle, serum metabolites and genetic risks, and the method is obviously superior to a prediction method only depending on clinical characteristics or genetic information.

Description

Technical Field

[0001] This invention relates to the field of disease risk prediction and precision medicine technology, and in particular to a method for establishing a risk prediction model for late-onset psoriasis and its application. Background Technology

[0002] Psoriasis is a chronic inflammatory skin disease that can be divided into two subtypes: early-onset and late-onset. Late-onset psoriasis (LOP) usually develops after the age of 40 and is closely related to lifestyle, environmental exposure, and metabolic status. It is often accompanied by metabolic diseases such as obesity, hypertension, and diabetes.

[0003] Current methods for predicting LOP risk mainly rely on clinical characteristics and genetic susceptibility indicators, but these methods cannot effectively reflect an individual's dynamic metabolic state and have limited predictive accuracy.

[0004] For example, existing technology CN 117174322A discloses a method and model for predicting clinical risk factors in psoriatic arthritis. The method includes: 1) collecting patient information; 2) using the chi-square test and Mann-Whitney U test to determine significant differences; 3) determining effective factors: 3.1) processing the patient information in the training set using STATA software, selecting LESSO regression, and obtaining variables with p-values ​​<0.05; 3.2) performing multivariate and univariate logistic regression analysis on the patient information in the training set, obtaining variables with p-values ​​<0.05; comparing the variables in 3.1) and 3.2), the same variables are taken as effective factors; 4) using STATA software, establishing a PsA prediction model using the effective factors; 5) inputting the specific patient information to be predicted into the prediction model to obtain the patient's percentage risk of disease. This invention, through the above method and model, can accurately and quickly predict the patient's risk of disease. However, this model has the following limitations: 1. Limited applicability: This method only applies to psoriatic arthritis (PsA) and cannot be generalized to other subtypes (such as late-onset psoriasis LOP). It ignores the heterogeneity of psoriasis and lacks accurate risk prediction for different clinical subtypes. 2. Single predictor: It relies solely on clinical characteristics (such as age, BSA, history of uveitis, lesions), failing to reflect the patient's lifestyle factors or metabolic status; it cannot capture dynamic metabolic changes in the body, and the prediction results may be limited. 3. Lack of metabolomics and genetic information: Existing methods lack integration with genetic risk scores (GRS) or lifestyle data, resulting in insufficient information dimensions in the model. 4. Model construction relies on traditional statistics: It only uses LASSO regression + Logistic regression, which has limited capabilities when handling high-dimensional, nonlinear data. The lack of application of machine learning / deep learning methods may affect the generalization performance and accuracy of the prediction. 5. Limited predictive accuracy and clinical translation: Due to the limitation of predictor factors, the sensitivity and specificity of the model are difficult to further improve. It can only be used for risk percentage estimation and lacks support for dynamic monitoring and individualized intervention.

[0005] Serum metabolomics can detect a variety of small molecule metabolites in the circulatory system through high-throughput technology, comprehensively reflecting the body's metabolic level. However, there is currently a lack of existing technology to integrate lifestyle information, serum metabolites, and genetic risk to construct predictive models. Summary of the Invention

[0006] The purpose of this invention is to provide a method for establishing a risk prediction model for late-onset psoriasis and its application.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A method for establishing a risk prediction model for late-onset psoriasis includes the following steps:

[0009] ① Collect lifestyle data, serum metabolite data, clinical characteristics, and polygenic risk scores from the subjects; randomly divide the subjects' data into training and validation sets;

[0010] ② Construct a healthy lifestyle score based on lifestyle data;

[0011] ③ Use multiple linear regression models to screen metabolites that are significantly associated with a healthy lifestyle;

[0012] ④ Use Cox regression model to screen metabolites that are significantly associated with the risk of late-onset psoriasis;

[0013] ⑤ Use an elastic network regression model to screen the set of key metabolites;

[0014] The key metabolite set includes R1-R11, where R1 is the unsaturation content, R2 is the albumin content, R3 is the cholesterol ester content in the largest HDL, R4 is the concentration of ultra-large HDL particles, R5 is the cholesterol content in the ultra-large HDL, R6 is the cholesterol ester content in the large HDL, R7 is the cholesterol content in the large HDL, R8 is the concentration of large HDL particles, R9 is the cholesterol ester content in HDL, R10 is the phenylalanine content, and R11 is the glycoprotein acetylation content.

[0015] ⑥ Input the sample features from the training set into the machine learning model XGBoost to obtain a risk prediction model for late-onset psoriasis; and evaluate the predictive ability of the model after testing it with a validation set.

[0016] In one preferred embodiment, the machine learning model is an extreme gradient boosting algorithm (XGBoost), which is a prior art technique.

[0017] In one preferred embodiment, the sample features of the machine learning model include: clinical information, a set of key metabolites, and a multigene risk score; or clinical information and a set of key metabolites.

[0018] In one preferred embodiment, clinical information includes sex, age, education level, alcohol consumption, race, TDI, BMI, smoking status, MET, and diet.

[0019] In one preferred embodiment, the data for the polygenic risk score are derived from 62 SNP loci identified in the psoriasis genome-wide association study (GWAS). The GWAS provides the sum of risk alleles for each variant locus, as well as the effect size between a single SNP and psoriasis. The formula for calculating the polygenic risk score is as follows: SNPs are the sum of risk alleles at each variant site, and the β coefficient is the effect size between a single SNP and psoriasis.

[0020] In one preferred embodiment, 80% of the subjects' data is randomly allocated to the training set, and the remaining 20% ​​is allocated to the validation set.

[0021] In one preferred embodiment, the inclusion criteria for late-onset psoriasis samples include any of the following: International Classification of Diseases (ICD-10) code L40 field 41270; or a clinical diagnosis of psoriasis; the late-onset psoriasis samples are from a group aged 37-73 years, and all newly reported psoriasis cases during the follow-up period occur after the age of 40.

[0022] The elastic network regression model is an existing technology. In a preferred embodiment, the formula for the elastic network regression model is as follows:

[0023]

[0024] Where n is the sample size, p is the number of covariate columns (61), λ is the regularization strength (0.006932275), α is the elastic network mixing parameter (α = 0.1), and β... j X represents the coefficients in the model; ij Let be the value of the j-th covariate for the i-th sample; the covariates are age, sex, race, education level, alcohol consumption status, Thomson quotient, hypertension, and baseline diabetes.

[0025] In one preferred embodiment, the covariates of the elastic network regression model are age, gender, race, education level, alcohol consumption status, Thomson criterion, baseline hypertension and diabetes. Substituting these covariates into the elastic network regression model yields the model coefficients.

[0026] The Cox regression model is an existing technology. In a preferred embodiment, the Cox regression model formula is as follows:

[0027] Where: h(t) is the instantaneous risk; h0(t) is the baseline risk, which changes over time; X i It is a metabolite; Z j β is a covariate. i and β j These are the regression coefficients for metabolites and covariates, respectively.

[0028] In one preferred embodiment, the independent variable of the Cox regression model is metabolite, and the covariates are age, sex, race, education level, alcohol consumption status, Thomson criterion, baseline hypertension and diabetes. Inputting the independent and covariates into the Cox regression model yields the instantaneous risk value.

[0029] The multiple linear regression model is an existing technology. In a preferred embodiment, the formula for the multiple linear regression model is as follows:

[0030] Where Y is the lifestyle score, Xj is the primary independent variable, β0 is the intercept, βj is the coefficient of the primary independent variable, Zk is the covariate, γk is the coefficient of the covariate, and ε is the error term; the primary independent variable is the primary candidate metabolite, and the covariates are gender, age, education level, race, alcohol consumption status, TDI, hypertension, and diabetes.

[0031] In one preferred embodiment, the primary independent variable of the multiple linear regression model is the primary candidate metabolite, and the covariates are gender, age, education level, race, alcohol consumption status, TDI, hypertension, and diabetes. Inputting the primary independent variable and covariates into the multiple linear regression model yields the lifestyle score.

[0032] In one preferred embodiment, the healthy lifestyle score includes dietary habits, body mass index (BMI), smoking status, and physical activity, with each behavioral item receiving 1 point;

[0033] The first item is BMI (kg / m²). 2 The ideal BMI range is defined as 18.5 kg / m², calculated by measuring weight and height. 2 ≤BMI<25kg / m 2 ;

[0034] The second point is that the ideal smoking condition is defined as a never-smoker.

[0035] Thirdly, regarding physical activity, the ideal state is defined as ≥150 minutes of moderate-intensity exercise, ≥75 minutes of high-intensity exercise, or ≥150 minutes of mixed exercise (moderate-intensity + high-intensity) per week.

[0036] The fourth item is a healthy eating pattern formulated with reference to the American Heart Association guidelines. The specific standards are: daily intake of ≥3 servings of fruit, ≥3 servings of vegetables, ≥2 servings of fish, and ≤1 serving of processed meat and red meat. If a participant meets at least 3 healthy eating patterns, they are considered to be in an ideal eating state.

[0037] Each ideal lifestyle factor is worth 1 point, with a total score ranging from 0 to 4 points. The higher the score, the healthier the lifestyle.

[0038] Based on the same inventive concept, the present invention also claims protection for the late-onset psoriasis risk prediction device obtained by the method for establishing the late-onset psoriasis risk prediction model.

[0039] Based on the same inventive concept, the present invention also claims protection for the application of the late-onset psoriasis risk prediction device in assessing the risk of late-onset psoriasis.

[0040] Based on the same inventive concept, this invention also claims protection for the application of the late-onset psoriasis risk prediction model obtained by the method described in this invention in evaluating the late-onset psoriasis risk prediction system.

[0041] Preferably, the sample features of the late-onset psoriasis risk prediction model include three types: clinical information, key metabolite set, and multigene risk score, or two types: clinical information and key metabolite set. The clinical features include gender, age, education level, alcohol consumption, race, TDI, BMI, smoking status, MET, and diet. The key metabolite set consists of the following 11 metabolites: unsaturation, albumin, cholesterol esters in ultra-large HDL, ultra-large HDL particle concentration, cholesterol in ultra-large HDL, cholesterol esters in large HDL, cholesterol in large HDL, large HDL particle concentration, cholesterol esters in HDL, phenylalanine, and glycoprotein acetylation.

[0042] The device or system for assessing the risk of late-onset psoriasis includes:

[0043] ① Data acquisition module, used to collect subjects' lifestyle data, serum metabolite data, clinical characteristics and multigene risk scores;

[0044] ②Lifestyle scoring module, used to construct a healthy lifestyle score;

[0045] ③ Metabolite screening module, used to screen key metabolites based on multiple linear regression, Cox regression and elastic network regression;

[0046] ④ The risk prediction module is used to input key metabolites, clinical characteristics, and multi-gene risk scores into the machine learning model XGBoost for risk prediction;

[0047] ⑤ Results output module, used to output individualized risk scores and risk levels.

[0048] Based on the same inventive concept, this invention also claims the use of a reagent for detecting the content of multiple metabolites in the preparation of reagents or kits for predicting psoriasis risk, wherein the content of the metabolites is the content of the following 11 metabolites: R1 is the unsaturation content, R2 is the albumin content, R3 is the cholesterol ester content in the maximum HDL, R4 is the concentration of ultra-large HDL particles, R5 is the cholesterol content in ultra-large HDL, R6 is the cholesterol ester content in large HDL, R7 is the cholesterol content in large HDL, R8 is the concentration of large HDL particles, R9 is the cholesterol ester content in HDL, R10 is the phenylalanine content, and R11 is the glycoprotein acetylation content.

[0049] Based on the same inventive concept, this invention also claims the use of a reagent for detecting the content of multiple metabolites in the preparation of reagents or kits for predicting psoriasis risk. The formula for predicting psoriasis risk is: Metabolic risk score = (-0.0598*R1) + (-0.0455*R2) + (-0.0132*R3) + (-0.0123*R4) + (-0.0087*R5) + (-0.0062*R6) + (-0.0026*R7) + (-0.0013*R8) + (-0.000 3*R9)+(0.0155*R10)+(0.0966*R11), where R1 is the unsaturation content, R2 is the albumin content, R3 is the cholesterol ester content in the largest HDL, R4 is the concentration of ultra-large HDL particles, R5 is the cholesterol content in the ultra-large HDL, R6 is the cholesterol ester content in the large HDL, R7 is the cholesterol content in the large HDL, R8 is the concentration of large HDL particles, R9 is the cholesterol ester content in HDL, R10 is the phenylalanine content, and R11 is the glycoprotein acetylation content.

[0050] Preferably, the unit for R1-R11 is mmol / L serum sample.

[0051] In one preferred embodiment, the metabolic risk score is standardized using z-scores (mean = 0, standard deviation = 1); the overall data is sorted from smallest to largest and divided into quintiles, with the smallest 20% of data being the first quintile and the largest 20% being the fifth quintile; the metabolic score is divided into three metabolic risk levels: the first quintile is good, the second to fourth quintiles are moderate, and the fifth quintile is poor.

[0052] In one preferred embodiment, the levels of multiple metabolites are detected by nuclear magnetic resonance.

[0053] The present invention will be further explained below:

[0054] To evaluate whether metabolomics biomarkers can be used to improve the prediction of late-onset psoriasis, this invention tested the following four models using the machine learning model XGBoost: (i) Model 1: basic clinical characteristics, including sex, age, education level, alcohol consumption, race, TDI, BMI, smoking status, MET, and diet; (ii) Model 2: Model 1 + Psoriasis Polygenic Risk Score (PRS); (iii) Model 3: Model 1 + significant metabolites; (iv) Model 4: Model 1 + Psoriasis Polygenic Risk Score (PRS) + significant metabolites (metabolites significantly associated with the incidence of late-onset psoriasis in this invention). The discriminative power of each model was evaluated in the total cohort population. Accuracy, specificity, sensitivity, F1 score, and area under the receiver operating characteristic curve (AUC) were used to assess model performance, and 95% confidence intervals were obtained through 1000 bootstrap samplings (percentile method). These analyses were performed using the scikit-learn and xgboost packages in Python.

[0055] Compared with the prior art, the beneficial effects of the present invention are:

[0056] 1) The model of this invention incorporates 11 serum metabolites, achieving high predictive accuracy with an AUC of 0.86, significantly outperforming prediction methods that rely solely on clinical characteristics (AUC = 0.78) and genetic information (AUC = 0.79). Improved predictive accuracy means earlier identification of high-risk individuals, enabling timely intervention.

[0057] 2) The model of this invention is highly operable: the required lifestyle information can be collected through a standard questionnaire; serum metabolite testing can be completed through routine physical examinations; and genetic information testing can be used long-term after a single test. This model does not require expensive or complex equipment, making it suitable for rapid application and promotion in community hospitals, primary health management institutions, or physical examination centers.

[0058] 3) The model of this invention can be dynamically monitored: the model can dynamically update the disease risk prediction based on the detection results at different time points. It supports long-term follow-up and intervention effect evaluation, for example, by periodically detecting changes in serum metabolites to assess the actual effect of lifestyle intervention.

[0059] 4) The model of this invention can realize personalized intervention guidance: it provides differentiated health intervention plans such as diet and exercise according to the risk level, so as to achieve precise health management. Attached Figure Description

[0060] Figure 1 This is a flowchart illustrating the process of screening and obtaining 11 key metabolites according to the present invention;

[0061] Figure 2 This is a volcano plot showing the results of the validation on the multiple linear regression validation set;

[0062] Figure 3 This describes the relationship between the metabolic risk score and the risk ratio (HR) for late-onset psoriasis in Model 3; where... Figure 3 A is a restricted cubic spline (RCS) plot that has been adjusted multiple times. Figure 3 B is a distribution chart of the cumulative incidence and risk ratio of psoriasis based on metabolic characteristics;

[0063] Figure 4 For each model, the receiver operating characteristic (ROC) curve and decision curve are provided; among them, Figure 4 A is the ROC curve. Figure 4 B represents the decision curve. Detailed Implementation

[0064] This invention is not limited to the specific embodiments listed below. Those skilled in the art can implement this invention using various other specific embodiments based on the content disclosed herein. Any modifications or alterations made to the design structure and concept of this invention fall within the protection scope of this invention. It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other.

[0065] Example 1

[0066] according to Figure 1 The flowchart is used to construct the model of this invention, as follows:

[0067] 1. Data Source

[0068] Between 2006 and 2010, over 500,000 participants aged 37 to 73 were recruited at 22 assessment centers in the UK. At baseline assessment, participants provided lifestyle and medical history information through electronic signatures, touchscreen questionnaires, interviews, and physical examinations. All participants donated blood, urine, and saliva samples for subsequent analysis and consented to follow-up health tracking through association with health-related records. The study was approved by the Northwest Multicenter Research Ethics Committee. Data from the prospective cohort study UK Biobank were utilized.

[0069] 2. Definition of Psoriasis

[0070] Individual medical record data were derived from the International Classification of Diseases (ICD-10). In addition, participants answered questions about physician-diagnosed diseases (self-reported diseases). In short, the definition of psoriasis included any of the following: ICD-10 code L40 (field 41270) and a self-reported clinical diagnosis of psoriasis (codes 1453 and 1477 in field 20002). Because this study only included adults to the elderly aged 37–73 years, and all newly reported psoriasis cases during the follow-up period occurred after age 40, it was classified as late-onset psoriasis (LOP). Follow-up continued until the date of the first psoriasis diagnosis, the end of follow-up (September 8, 2023), or the participant's death, whichever occurred first.

[0071] 3. Assessment of lifestyle factors and construction of a healthy lifestyle score

[0072] This invention constructs a healthy lifestyle scoring system based on four recognized risk factors for psoriasis: dietary habits, body mass index (BMI), smoking status, and physical activity. Participants are required to assign 1 point to each of the above behavioral items according to national recommended standards.

[0073] The first item is BMI (kg / m²). 2 The ideal BMI range is defined as 18.5 kg / m², calculated by measuring weight and height. 2 ≤BMI<25kg / m 2 .

[0074] The second point is that the ideal smoking condition is defined as a person who has never smoked.

[0075] Thirdly, regarding physical activity, the ideal state is defined as ≥150 minutes of moderate-intensity exercise, ≥75 minutes of high-intensity exercise, or ≥150 minutes of mixed exercise (moderate-intensity + high-intensity) per week.

[0076] Fourthly, the healthy eating pattern is based on the American Heart Association guidelines, specifically: daily intake of ≥3 servings of fruit, ≥3 servings of vegetables, ≥2 servings of fish, and ≤1 serving of processed meat and red meat. Participants who meet at least 3 healthy eating patterns are considered to be in an ideal eating state.

[0077] Each ideal lifestyle factor is worth 1 point, with a total score ranging from 0 to 4 points. The higher the score, the healthier the lifestyle.

[0078] 4. Covariates

[0079] Several potential confounding factors were adjusted for in these models, including age (continuous), sex (male, female), race (white, non-white, or missing values), TDI score (continuous), alcohol consumption status (never drank, used to drink, currently drink), education level (university degree, professional qualification, advanced education certificate or equivalent, general secondary education certificate or equivalent, or missing values), and prevalence of hypertension and diabetes (yes or no). Information on cholesterol, blood pressure, and blood sugar medication use was obtained through interviews. The inclusion criteria for cholesterol, blood pressure, and blood sugar medication use are shown in Table 1. The degree of social deprivation was determined using the Townsend Deprivation Index (TDI), which includes four dimensions: unemployment rate, car ownership rate, household overcrowding, and homeowner occupation. Data was sourced from the national census based on individual postal codes. A higher TDI score indicates a greater degree of social deprivation. Notably, participants with missing categories for lifestyle factors were excluded.

[0080] Table 1. Field IDs and ICD codes used by the UK Biobank to define baseline diseases or medications.

[0081]

[0082] Note: Self-reported medical history and medication information are used only to define current conditions. Undiagnosed current type 2 diabetes cases are determined by fasting blood glucose (≥11.1 mmol / L) or glycated hemoglobin (≥48 mmol / mol [6.5%]). Undiagnosed current hypertension cases are determined by systolic blood pressure (≥140 mmHg) or diastolic blood pressure (≥90 mmHg).

[0083] 5. Detection of metabolic biomarkers

[0084] Researchers used a high-throughput targeted nuclear magnetic resonance (NMR) metabolomics platform to randomly select non-fasting serum samples from 252,054 subjects at baseline. This platform can simultaneously detect 249 metabolic biomarkers (168 absolute levels, mmol / L, and 81 ratios), covering lipids, fatty acids, apolipoproteins, amino acids, ketone bodies, and glycolytic metabolites. Considering the potential for collinearity in metabolomics data, this invention excluded ratios and ultimately used only 168 absolute levels of metabolites for analysis: total cholesterol, total cholesterol minus high-density lipoprotein cholesterol, residual cholesterol, very low-density lipoprotein cholesterol (VLDL-C), clinical low-density lipoprotein cholesterol (LDL-C), low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, total triglycerides, triglycerides in VLDL, triglycerides in low-density lipoprotein, triglycerides in high-density lipoprotein, total phospholipids in lipoprotein particles, phospholipids in VLDL, and low-density lipoprotein. Phospholipids in lipoproteins, phospholipids in high-density lipoprotein (HDL), total esterified cholesterol, cholesterol esters in very low-density lipoprotein (VLDL), cholesterol esters in low-density lipoprotein (LDL), cholesterol esters in high-density lipoprotein (HDL), total free cholesterol, free cholesterol in VLDL, free cholesterol in LDL, free cholesterol in HDL, total lipids in lipoprotein particles, total lipids in VLDL, total lipids in LDL, total lipids in HDL, total concentration of lipoprotein particles, concentration of VLDL particles, concentration of LDL particles, concentration of HDL particles, VLDL... Mean diameter of protein particles, mean diameter of low-density lipoprotein particles, mean diameter of high-density lipoprotein particles, glycerol phosphate, total choline, phosphatidylcholine, sphingomyelin, apolipoprotein B, apolipoprotein A1, total fatty acids, degree of unsaturation, omega-3 fatty acids, omega-6 fatty acids, polyunsaturated fatty acids, monounsaturated fatty acids, saturated fatty acids, linoleic acid, docosahexaenoic acid (DHA), alanine, glutamine, glycine, histidine, total concentration of branched-chain amino acids, isoleucine, leucine, valine, phenylalanine, tyrosine, glucose, lactic acid, pyruvic acid, citric acid, 3-hydroxybutyric acid, acetic acid, acetoacetic acid, acetone, muscle Anhydride, albumin, glycoprotein acetyl groups, concentrations of chylomicrons and very large and very low density lipoproteins (VLDL), total lipids in chylomicrons and VLDL, phospholipids in chylomicrons and VLDL, cholesterol in chylomicrons and VLDL, cholesterol esters in chylomicrons and VLDL, free cholesterol in chylomicrons and VLDL, triglycerides in chylomicrons and VLDL, concentration of very large and very low density lipoproteins (VLDL), total lipids in VLDL, phospholipids in VLDL, and cholesterol in VLDL.Cholesterol esters in very large and very low density lipoprotein (VLDL), free cholesterol in very large and very low density lipoprotein (VLDL), triglycerides in very large and very low density lipoprotein (VLDL), concentration of large and very low density lipoprotein (VLDL) particles, total lipids in large and very low density lipoprotein (VLDL), phospholipids in large and very low density lipoprotein (VLDL), cholesterol in large and very low density lipoprotein (VLDL), cholesterol esters in large and very low density lipoprotein (VLDL), free cholesterol in large and very low density lipoprotein (VLDL), triglycerides in large and very low density lipoprotein (VLDL), concentration of intermediate-low density lipoprotein (ILL) particles, total lipids in intermediate-low density lipoprotein (ILL), phospholipids in intermediate-low density lipoprotein (ILL), cholesterol in intermediate-low density lipoprotein (ILL), cholesterol esters in intermediate-low density lipoprotein (ILL), free cholesterol in intermediate-low density lipoprotein (ILL), intermediate-low density lipoprotein (ILL) particles, concentration of intermediate-low density lipoprotein (ILL), ... Triglycerides in protein, concentration of small low-density lipoprotein (LDL) particles, total lipids in small low-density lipoprotein (LDL), phospholipids in small low-density lipoprotein (LDL), cholesterol in small low-density lipoprotein (LDL), cholesterol esters in small low-density lipoprotein (LDL), free cholesterol in small low-density lipoprotein (LDL), triglycerides in small low-density lipoprotein (LDL), concentration of very low-density lipoprotein (VLDL) particles, total lipids in VLDL, phospholipids in VLDL, cholesterol in VLDL, cholesterol esters in VLDL, free cholesterol in VLDL, triglycerides in VLDL, concentration of intermediate-density lipoprotein (IDL) particles, total lipids in intermediate-density lipoprotein (IDL), intermediate... Phospholipids in LDL, cholesterol in MDL, cholesterol esters in MDL, free cholesterol in MDL, triglycerides in MDL, LDL particle concentration, total lipids in LDL, phospholipids in LDL, cholesterol in LDL, cholesterol esters in LDL, free cholesterol in LDL, triglycerides in LDL, LDL particle concentration, total lipids in LDL, phospholipids in LDL, cholesterol in LDL, cholesterol esters in LDL, free cholesterol in LDL, triglycerides in LDL, LDL particle concentration, total lipids in LDL, phospholipids in LDL, cholesterol in LDL, cholesterol esters in LDL, free cholesterol in LDL, and triglycerides in LDL. Triglycerides, low-density lipoprotein (LDL) particle concentration, total lipids in LDL, phospholipids in LDL, cholesterol in LDL, cholesterol esters in LDL, free cholesterol in LDL, triglycerides in LDL, high-density lipoprotein (HDL) particle concentration, total lipids in HDL, phospholipids in HDL, cholesterol in HDL, cholesterol esters in HDL, free cholesterol in HDL, triglycerides in HDL, high-density lipoprotein (HDL) particle concentration, total lipids in HDL, phospholipids in HDL, cholesterol in HDL, cholesterol esters in HDL, free cholesterol in HDL, triglycerides in HDL, high-density lipoprotein (HDL) particle concentration, total lipids in HDL, phospholipids in HDL, cholesterol in HDLCholesterol esters in large high-density lipoprotein (HDL), free cholesterol in large HDL, triglycerides in large HDL, concentration of intermediate-high-density lipoprotein (MDL) particles, total lipids in MDL, phospholipids in MDL, cholesterol in MDL, cholesterol esters in MDL, free cholesterol in MDL, triglycerides in MDL, concentration of small high-density lipoprotein (HDL), total lipids in small HDL, phospholipids in small HDL, cholesterol in small HDL, cholesterol esters in small HDL, free cholesterol in small HDL, triglycerides in small HDL.

[0085] 6. Multiple linear regression analysis

[0086] Multiple linear regression analysis was employed to identify and validate lifestyle-related metabolites. This invention first excluded individuals without lifestyle factors from the original 252,054 participants with serum metabolomics data, screening for metabolic characteristics of healthy lifestyles from the 190,692 participants with complete data. Subsequently, 3,880 participants with baseline psoriasis were selected as the case-control sample, and matched with an equal number of control groups (with consistent age, sex, education level, race, Townsend Deprivation Index (TDI) score, alcohol consumption, BMI, physical activity, dietary habits, smoking status, baseline hypertension, and diabetes, etc.) to optimize statistical power.

[0087] Multiple linear regression model construction: A multiple linear regression model was constructed for each candidate metabolite, using that metabolite as the primary predictor, while adjusting for covariates such as gender, age, education level, race, alcohol consumption status, TDI, hypertension, and diabetes. The model format is as follows:

[0088]

[0089] Where Y is the lifestyle score, Xj is the primary independent variable, β0 is the intercept, βj is the coefficient of the primary independent variable, Zk is the covariate, γk is the covariate coefficient, and ε is the error term.

[0090] The primary independent variable was the major candidate metabolite, and the covariates were sex, age, education level, race, alcohol consumption status, total intradermal drug intake (TDI), hypertension, and diabetes. The intercepts and coefficients for each metabolite are as follows:

[0091] Table 2 Intercepts and coefficients for each metabolite

[0092]

[0093]

[0094]

[0095]

[0096]

[0097]

[0098] Table 3 Coefficients of each covariate

[0099]

[0100]

[0101]

[0102]

[0103]

[0104] The error term (residual / ε) is the random residual hidden in the model, used to reflect the deviation between the observed values ​​and the model's predictions.

[0105] Results Summary: Statistical analysis for significance levels (p-values) was performed using R Studio software. The principal independent variable coefficients and significance levels for each variable were extracted to form a complete results table, used to assess the impact of each metabolite on the healthy lifestyle score. Bonfferoni adjustment was used for significance levels, and metabolites with p-adjusted < 0.05 were defined as metabolites associated with a healthy lifestyle. Results are shown in Tables 4-7 and... Figure 2 As shown.

[0106] Table 4 lists the baseline characteristics of participants in the UK Biobank study (n=7760) based on the clustered healthy lifestyle scores.

[0107]

[0108]

[0109] Data are expressed as median ± interquartile range (IQR) and sample size (%). A p-value < 0.05 was considered statistically significant.

[0110] Abbreviations: BMI, Body Mass Index (calculated by dividing weight (kg) by the square of height (m); SD, Standard Deviation; UK Biobank, UK Biobank.

[0111] Table 5 identifies metabolites associated with lifestyle.

[0112]

[0113]

[0114]

[0115]

[0116] Bonferroni adjustment was used to correct the p-value for multiple tests.

[0117] Next, after excluding baseline psoriasis patients and their matched participants, the association between lifestyle and metabolites in the validation set was further validated (n = 182,932).

[0118] Table 6 lists the baseline characteristics of UK Biobank study participants in the validation set based on healthy lifestyle scores (n = 182,932).

[0119]

[0120]

[0121] Data are expressed as median ± interquartile range (IQR) and sample size (%). A p-value < 0.05 was considered statistically significant.

[0122] Abbreviations: BMI, Body Mass Index (calculated by dividing weight (kg) by the square of height (m); SD, Standard Deviation; UK Biobank, UK Biobank.

[0123] Table 7. Lifestyle-related metabolites in the validation set

[0124]

[0125]

[0126]

[0127]

[0128]

[0129] Bonferroni adjustment was used to correct the p-value for multiple tests.

[0130] Baseline characteristics of participants based on their healthy lifestyle scores are detailed in Tables 4 and 6. In the discovery set, the mean age at baseline was 56.67 years, and the mean follow-up time was 14.57 years. Notably, only 27.6% of participants met the ideal body mass index (BMI). Furthermore, 78.2% met the ideal metabolic equivalent (MEI) criteria, and 20.1% maintained ideal dietary habits. 44.7% were non-smokers, and the baseline prevalence of hypertension and diabetes was 28.3% and 6.3%, respectively. The validation set showed similar trends to the discovery set in terms of healthy lifestyle scores and behavioral patterns, with only minor differences.

[0131] In the discovery set analysis, 134 out of 168 metabolites were identified as being associated with healthy lifestyle scores (p-adjusted < 0.05, Table 5). Visualizing the correlation between these metabolites and healthy lifestyle scores using a volcano plot revealed that only 37 metabolites showed a positive correlation, while the rest showed a negative correlation. Among these, triglycerides, glycoprotein acetylation, and various lipid metabolites associated with very low-density lipoprotein (VLDL) showed the most significant negative correlations. Conversely, unsaturation associated with high-density lipoprotein (HDL) and various lipid metabolites showed the strongest positive correlations. Notably, the results in the validation set remained stable among these 134 statistically significant metabolites (p-adjusted < 0.05, see Table 7).

[0132] 7. A sequential analysis strategy combining Cox regression and elastic network regression was used to screen metabolites that were significantly associated with the risk of late-onset psoriasis.

[0133] 7.1 Using the Cox regression model to select psoriasis-related metabolites

[0134] To assess the association between lifestyle-related metabolites and the incidence of late-onset psoriasis, researchers performed a single metabolite fit analysis of a Cox regression model on 186,812 participants (whose information is summarized in the table below) after excluding patients with pre-existing psoriasis.

[0135] Table 8. Descriptive characteristics of LOP incidence rates in the UK Biobank study (n = 186,812)

[0136]

[0137]

[0138] Model Construction and Analysis: For each metabolite in the candidate metabolite set, a multivariate Cox proportional hazards regression model was constructed. The transient risk h(t) was defined as 1 for transient onset and 0 for no onset. This metabolite was used as the primary dependent variable, while adjusting for covariates (age, sex, race, education level, alcohol consumption status, Thomson Scale index, baseline hypertension, and diabetes). The model form is as follows:

[0139]

[0140] Where: h(t) is the instantaneous risk; h0(t) is the baseline risk, which changes over time; X i It is a metabolite; Z j β is a covariate. i and β j These are the regression coefficients for metabolites and covariates, respectively, where the regression coefficient for metabolites is defined as HR.

[0141] The baseline characteristics of participants in the cohort analysis are summarized in Table 9.

[0142] Table 9. Relationship between lifestyle-related metabolites and LOP in UK biobank research.

[0143]

[0144]

[0145] Table 9 shows the association between lifestyle-related metabolites and the subsequent incidence of late-onset psoriasis. The results showed that among the 134 lifestyle-related metabolites, 52 were significantly associated with the incidence of late-onset psoriasis after multivariate adjustment (p adjusted < 0.05). Of these, 22 metabolites showed a negative association with the risk of late-onset psoriasis, while the remaining 30 showed a positive association.

[0146] 7.2 Reducing the effective characteristics of psoriasis-related metabolites using an elastic network model

[0147] An EN-regularized COX-PH model using an elastic network was optimized for the 52 candidate metabolites, and trained with 10-fold cross-validation through an internal test cohort. Covariates included age, sex, race, education level, alcohol consumption status, Thomson criterion, baseline hypertension, and diabetes. The formula is as follows:

[0148]

[0149] in, Here, n is the model coefficient, p is the number of covariate columns (61), λ is the regularization strength (0.006932275), α is the elastic network mixing parameter (α = 0.1), and β is the model coefficient. jX represents the coefficients in the model; ij This represents the value of the j-th covariate for the i-th sample. Covariates include age, sex, race, education level, alcohol consumption, Thomson Scale index, baseline hypertension, and diabetes. If the covariate is age, the covariate value is the specific age. If the covariate is hypertension, the covariate value is 1 for those with hypertension and 0 for those without. The results are shown in Table 10.

[0150] Table 10 Metabolites with non-zero coefficients after elastic network regression

[0151] Metabolites (n=11) coefficient Unsaturation -0.0598 albumin -0.0455 Cholesterol esters in ultra-large HDL -0.0132 Ultra-high HDL particle concentration -0.0123 Cholesterol in Ultra HDL -0.0087 Cholesterol esters in large HDL -0.0062 Cholesterol in HDL -0.0026 Large HDL particle concentration -0.0013 Cholesterol esters in HDL -0.0003 Phenylalanine 0.0155 Glycoprotein acetylation 0.0966

[0152] The results showed that 11 metabolites were ultimately associated, including 7 lipoprotein subclasses, 2 ketone bodies, unsaturation, and phenylalanine.

[0153] 8. Metabolic risk score and metabolic characteristics

[0154] To further explore the comprehensive impact of important metabolites on the risk of late-onset psoriasis, this invention constructs a metabolic risk scoring system based on metabolites with non-zero coefficients. The score is calculated using coefficients obtained from a regularized Cox regression model with an elastic network as weights. The metabolic risk score is calculated as follows: Metabolic Risk Score = (-0.0598*R1) + (-0.0455*R2) + (-0.0132*R3) + (-0.0123*R4) + (-0.0087*R5) + (-0.0062*R6) + (-0.0026*R7) + (-0.0013*R8) + (-0.0003*R9) + (0.0155*R10) + (0.0966*R11). In the scoring system, R1 represents the unsaturation level, R2 represents the albumin level, R3 represents the cholesterol ester content in the largest HDL particles, R4 represents the concentration of ultra-large HDL particles, R5 represents the cholesterol content in the ultra-large HDL particles, R6 represents the cholesterol ester content in the large HDL particles, R7 represents the cholesterol content in the large HDL particles, R8 represents the concentration of large HDL particles, R9 represents the cholesterol ester content in HDL particles, R10 represents the phenylalanine level, and R11 represents the glycoprotein acetylation level. The weighting symbols (negative values ​​represent protective factors, positive values ​​represent harmful factors) reflect the direction of the metabolite's effect in the scoring system, with higher scores indicating poorer metabolic status. Scores are standardized using z-scores (mean = 0, standard deviation = 1). The overall data are sorted from smallest to largest and divided into quintiles, with the smallest 20% of data forming the first quintile and the largest 20% forming the fifth quintile. The metabolic score was divided into three metabolic risk levels: good (1st quintile), moderate (2nd to 4th quintiles), and poor (5th quintile).

[0155] We continued to use the Cox proportional hazards model to calculate the hazard ratio (HR) and its 95% confidence interval (CI) between metabolic status (metabolic risk score and metabolic characteristics) and the risk of late-onset psoriasis.

[0156] The model is in the following form:

[0157] h(t) = h0(t) · exp(βiXi + ∑ j βjZj);

[0158] Where: h(t) is the instantaneous risk; h0(t) is the baseline risk, which changes over time; X i Healthy lifestyle score / metabolic risk score / metabolic characteristics; Z j β is a covariate. i and β j These are the regression coefficients for metabolites and covariates, respectively, where the regression coefficient for metabolites is defined as HR.

[0159] Cox Model 1 uses age and sex as covariates; Cox Model 2 is multivariate adjusted and includes age, sex, education level, race, alcohol consumption, and total diabetes index (TDI) as covariates; Cox Model 3 is a complete model and includes age, sex, education level, race, alcohol consumption, total diabetes index (TDI), cholesterol-lowering medication use, baseline diabetes, and hypertension as covariates.

[0160] The results of the complete model are as follows Figure 3 As shown. Among them, Figure 3 A multi-adjusted restricted cubic spline (RCS) plot shows the relationship between metabolic risk score (in standard deviation) and the risk ratio (HR) for late-onset psoriasis. The blue curve represents the hazard function, and the blue shaded area represents the 95% confidence interval. The median was used as the reference after excluding the top 2.5% of the distribution range. The spline curve has been adjusted for variables such as age, sex, education level, race, alcohol consumption, TDI (Total Disease Index), cholesterol-lowering medication use, baseline diabetes, and hypertension. Figure 3 B represents the distribution of cumulative incidence and risk ratios of psoriasis based on metabolic characteristics.

[0161] The results showed a linear positive correlation between metabolic score and the risk of late-onset psoriasis. Figure 3 A). Participants with poor metabolic characteristics had a more than twice the risk compared to those with good metabolic characteristics (HR 2.30, 95% CI 1.98–2.68, P < 0.001). Figure 3 B)

[0162] The hazard ratios (HRs) and their 95% confidence intervals (CIs) between metabolic status (metabolic risk score and metabolic characteristics) and the risk of late-onset psoriasis in three different models are shown in Table 11.

[0163] Table 11 Unadjusted and multi-adjusted HRs (n = 186,812) for the risk of late-onset psoriasis associated with healthy lifestyle scores and metabolic status.

[0164]

[0165] Model 1: Adjusted based on age and gender.

[0166] Model 2: Adjusted for age, gender, education level, race, drinking status, and TDI.

[0167] Model 3: Model 2+ uses cholesterol-lowering drugs, baseline diabetes, and hypertension.

[0168] The results show that after multiple adjustments to the three models, the HR value did not change significantly, and the P value remained significant, indicating that the results of the 11 screened metabolites are reliable.

[0169] 9. Polygenic Risk Score

[0170] Multigene risk scoring for psoriasis was performed using the Psoriasis Multigene Risk Score (PRS). Data for the Psoriasis Multigene Risk Score (PRS) were derived from 62 SNP loci identified in the Genome-wide Association Study of Psoriasis (GWAS), and only SNPs available in this database were selected (Bycroft C, Freeman C, Petkova D, Band G, Elliott LT, Sharp K, et al. The UK Biobank resource with deep phenotyping and genomic data. Nature 2018; 562(7726):203-209). The GWAS provides the sum of risk alleles for each variant locus, as well as the effect size between a single SNP and psoriasis. The PRS calculation formula is as follows: Where SNP represents the sum of risk alleles at each variant site, and β coefficient is the effect size between a single SNP extracted from the above GWAS study and psoriasis.

[0171] 10. Construct a model to predict late-onset psoriasis events.

[0172] An optimal combination model for predicting late-onset psoriasis events was constructed using the Extreme Gradient Boosting (XGBoost) algorithm, and these models were optimized and compared. Sample characteristics included: basic clinical features, polygenic risk scores, and significant metabolites (metabolites significantly associated with the onset of late-onset psoriasis in this invention). Basic clinical features included sex, age, education level, alcohol consumption, race, TDI, BMI, smoking status, MET, and diet. Significant metabolites included R1-R11, where R1 is the unsaturation content, R2 is the albumin content, R3 is the cholesterol ester content in the largest HDL, R4 is the concentration of ultra-large HDL particles, R5 is the cholesterol content in the ultra-large HDL, R6 is the cholesterol ester content in the large HDL, R7 is the cholesterol content in the large HDL, R8 is the concentration of large HDL particles, R9 is the cholesterol ester content in HDL, R10 is the phenylalanine content, and R11 is the glycoprotein acetylation content.

[0173] These calculations were performed using the scikit-learn and xgboost model packages in Python software (both existing technologies). The model parameters are as follows: the learning task is set as binary classification "binary:logistic"; number of trees (n_estimators): 50; maximum depth of each tree (max_depth): 6; lower bound of the sum of sample weights required for a leaf node (min_child_weight): 3; minimum loss reduction requirement for splits (gamma): 0.8; proportion of feature subsets used by each tree during splits (colsample_bytree): 0.5; L1 regularization coefficient (alpha): 0.6; L2 regularization coefficient (lambda): 0.2; learning rate (learning_rate): 0.1.

[0174] Training process:

[0175] To evaluate whether metabolomics biomarkers can be used to improve the prediction of late-onset psoriasis, this invention tested the following four models using the machine learning model XGBoost: (i) Model 1: basic clinical characteristics, including sex, age, education level, alcohol consumption, ethnicity, TDI, BMI, smoking status, MET, and diet; (ii) Model 2: Model 1 + Psoriasis Polygenic Risk Score (PRS); (iii) Model 3: Model 1 + significant metabolites; (iv) Model 4: Model 1 + Psoriasis Polygenic Risk Score (PRS) + significant metabolites (metabolites significantly associated with the onset of late-onset psoriasis in this invention). These data were obtained from the training set. This invention evaluated the discriminative power of each model in the total cohort population, using accuracy, specificity, sensitivity, F1 score, and area under the receiver operating characteristic curve (AUC) to assess model performance, and obtaining corresponding 95% confidence intervals through 1000 bootstrap samplings (percentile method). These analyses were performed using the scikit-learn and xgboost packages in Python.

[0176] Table 12 Forecast Explanation

[0177] Actual / Forecast Positive Negative class Positive class (P) TP (True Example) FN (false negatives) Negative class (N) FP (False Positive Example) TN (true negative example)

[0178] TP: Actually positive class, predicted to be positive class

[0179] TN: Actually a negative class, predicted as a negative class

[0180] FP: Actually a negative class, predicted as a positive class

[0181] FN: Actually a positive class, predicted as a negative class

[0182] Accuracy:

[0183]

[0184] This indicates the percentage of correct predictions overall.

[0185] Sensitivity (Recall / True Positive Rate):

[0186]

[0187] This indicates the ability to identify positive samples.

[0188] Specificity (True Negative Rate):

[0189]

[0190] This indicates the ability to identify negative samples.

[0191] Precision:

[0192]

[0193] This indicates the proportion of samples predicted as positive that are actually positive.

[0194] F1 score:

[0195]

[0196] That is, the harmonic mean of precision and recall.

[0197] Area Under the ROC Curve (AUC): The horizontal axis of the ROC curve is 1-Specificity, and the vertical axis is Sensitivity.

[0198] AUC is defined as:

[0199]

[0200] Among them, TPR is the sensitivity. FPR is the false positive rate. In practical calculations, AUC is typically calculated by numerically integrating the area under the ROC curve using the trapezoidal rule. The results are as follows: Figure 4 As shown in Table 13. Figure 4 A is the ROC curve. Figure 4 B represents the decision curve.

[0201] Table 13 Model Evaluation

[0202] result Model 1 Model 2 Model 3 Model 4 AUC 0.789(0.786-0.792) 0.797(0.794-0.801) 0.851(0.848-0.854) 0.860(0.857-0.863) accuracy 0.717(0.714-0.721) 0.722(0.719-0.726) 0.771(0.768-0.774) 0.781(0.778-0.784) Specificity 0.685(0.681-0.691) 0.692(0.688-0.697) 0.740(0.736-0.745) 0.750(0.745-0.754) Sensitivity 0.749(0.745-0.754) 0.753(0.748-0.757) 0.802(0.798-0.806) 0.812(0.808-0.817) F1 score 0.726(0.723-0.730) 0.731(0.727-0.734) 0.777(0.774-0.781) 0.788(0.785-0.791)

[0203] Figure 4 The receiver operating characteristic (ROC) curves and decision curves for each model are presented. Incorporating metabolites into models including clinical features, and into models combining clinical features with the PRS (Personal Risk Syndrome), significantly improved predictive performance, with AUC values ​​reaching 0.851 (95% confidence interval: 0.848–0.854) and 0.860 (95% confidence interval: 0.857–0.863), respectively. Notably, the combined model integrating metabolites into the traditional clinical feature and genetic risk model showed the best performance across all metrics, including accuracy, sensitivity, specificity, F1 score, and net clinical benefit.

[0204] Example 2

[0205] Internal validation:

[0206] The model is the same as in Example 1. Participants were divided into training, testing, and internal validation sets in a 7:2:1 ratio. The Synthetic Minority Oversampling Technique (SMOTE) preprocessing algorithm was used to balance the ratio of target to non-target samples. Hyperparameter settings were optimized using a grid search method with five-fold cross-validation. The model performance was evaluated using metrics such as accuracy, specificity, sensitivity, F1 score, and area under the receiver operating characteristic (AUC). The results are shown in the table below.

[0207] Table 14 Internal Validation Results

[0208] result Model 1 Model 2 Model 3 Model 4 AUC 0.792 0.843 0.874 0.886 accuracy 0.705 0.727 0.768 0.773 Specificity 0.704 0.726 0.764 0.773 Sensitivity 0.754 0.843 0.862 0.894 F1 score 0.736 0.797 0.822 0.844

[0209] Model validation showed that the combined model, which integrates metabolites into the traditional clinical features and genetic risk model, performed best in terms of accuracy, sensitivity, specificity, and F1 score.

[0210] It should be noted that the above embodiments are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is impossible to exhaustively list all possible implementations here. All obvious variations or modifications derived from the technical solutions of this invention are still within the scope of protection of this invention.

Claims

1. A method of establishing a late-onset psoriasis risk prediction model, characterized by, The method comprises the following steps: ① Collecting lifestyle data, serum metabolite data, clinical characteristics and polygenic risk score of the subjects; dividing the data of the subjects into a training set and a validation set at random; ② Constructing a healthy lifestyle score according to the lifestyle data; ③ Screening metabolites significantly related to the healthy lifestyle by using a multiple linear regression model; ④ Screening metabolites significantly related to the risk of late-onset psoriasis by using a Cox regression model; ⑤ Screening a set of key metabolites by using an elastic network regression model; The set of key metabolites comprises R1-R11, wherein R1 is unsaturation content, R2 is albumin content, R3 is cholesteryl ester content in very large HDL, R4 is very large HDL particle concentration content, R5 is cholesteryl content in very large HDL, R6 is cholesteryl ester content in large HDL, R7 is cholesteryl content in large HDL, R8 is large HDL particle concentration content, R9 is cholesteryl ester content in HDL, R10 is phenylalanine content, and R11 is glycoprotein acetylation content; ⑥ Inputting sample characteristics in the training set into a machine learning model XGBoost to obtain a late-onset psoriasis risk prediction model; and testing and evaluating the prediction ability of the model by using the validation set; The sample characteristics comprise three of clinical characteristics, the set of key metabolites and the polygenic risk score, or two of the clinical characteristics and the set of key metabolites.

2. The establishment method according to claim 1, characterized by, The clinical characteristics comprise gender, age, education level, drinking status, race, TDI, BMI, smoking status, MET and diet.

3. The establishment method of claim 1, wherein, The data of polygenic risk score was derived from 62 SNPs identified in the genome-wide association study of psoriasis, the sum of risk alleles of each variant site provided in the genome-wide association study of psoriasis, and the effect size between single SNP and psoriasis; the calculation formula of polygenic risk score is as follows: Wherein SNP is the sum of risk alleles of each variant site, and β coefficient is the effect size between single SNP and psoriasis.

4. The establishment method of claim 1, wherein, 80% of the data of the subjects is divided into the training set at random, and the remaining 20% is divided into the validation set.

5. A late-onset psoriasis risk prediction device obtained by the establishment method according to any one of claims 1-4.

6. Application of the late-onset psoriasis risk prediction model obtained by the establishment method according to any one of claims 1-4 in evaluating a late-onset psoriasis risk prediction system.

7. Use according to claim 6, characterized in that, The sample characteristics of the late-onset psoriasis risk prediction model comprise three of clinical information, the set of key metabolites and the polygenic risk score, or two of the clinical information and the set of key metabolites; the clinical characteristics comprise gender, age, education level, drinking status, race, TDI, BMI, smoking status, MET and diet; and the set of key metabolites comprises the following 11 metabolites: unsaturation, albumin, cholesteryl ester in very large HDL, very large HDL particle concentration, cholesteryl in very large HDL, cholesteryl ester in large HDL, cholesteryl in large HDL, large HDL particle concentration, cholesteryl ester in HDL, phenylalanine and glycoprotein acetylation.

8. Use of a reagent for detecting the content of a plurality of metabolites in the manufacture of a reagent or kit for predicting the risk of psoriasis, characterized in that, The metabolite content comprises the following 11 metabolite contents: unsaturation, albumin, cholesteryl ester in very large HDL, very large HDL particle concentration, cholesteryl in very large HDL, cholesteryl ester in large HDL, cholesteryl in large HDL, large HDL particle concentration, cholesteryl ester in HDL, phenylalanine and glycoprotein acetylation.

9. Use according to claim 8, characterized in that, The calculation formula for predicting the risk of psoriasis by detecting the content of 11 metabolites is: metabolic risk score = (-0.0598*R1) + (-0.0455*R2) + (-0.0132*R3) + (-0.0123*R4) + (-0.0087*R5) + (-0.0062*R6) + (-0.0026*R7) + (-0.0013*R8) + (-0.0003*R9) + (0.0155*R10) + (0.0966*R11), wherein R1 is the content of unsaturation, R2 is the content of albumin, R3 is the content of cholesterol ester in very large HDL, R4 is the content of very large HDL particle concentration, R5 is the content of cholesterol in very large HDL, R6 is the content of cholesterol ester in large HDL, R7 is the content of cholesterol in large HDL, R8 is the content of large HDL particle concentration, R9 is the content of cholesterol ester in HDL, R10 is the content of phenylalanine, and R11 is the content of glycoprotein acetylation.

10. Use according to claim 9, characterized in that, The metabolic risk score is standardized by z-score, with mean = 0 and standard deviation = 1; the overall data is sorted from small to large, and divided into five quintiles, i.e. the smallest 20% of data is the first quintile, and the largest 20% of data is the fifth quintile; the metabolic score is divided into three types of metabolic risk levels: the first quintile is good type, the second to fourth quintiles are moderate type, and the fifth quintile is suboptimal type.

Images