Gout recurrence prediction model and construction method thereof

Abstract

The invention relates to the technical field of prediction models, and particularly discloses a gout recurrence prediction model and a construction method thereof. Clinical index data and the standardized protein expression level of immune-related molecules in plasma are used for constructing the gout recurrence prediction model, the gout recurrence prediction model obtained through the method has the stable prediction capacity, and MMP1 achieves the very high AUC value and accuracy and the lowest Brier score.

Description

Technical Field

[0001] The present application relates to the technical field of prediction models, and in particular to a gout recurrence prediction model and a method for constructing the same. Background Art

[0002] Gouty arthritis (GA) is a chronic inflammatory disease characterized by hyperuricemia and the accumulation of monosodium urate (MSU) crystals in joints and soft tissues. The disease presents with acute attacks characterized by severe pain and swelling. In the late stages, it progresses to chronic gouty nodules accompanied by persistent inflammation, joint destruction, and high disability.

[0003] Despite the effectiveness of urate-lowering treatments (ULTs) and anti-inflammatory drugs in controlling hyperuricemia and acute inflammation, the clinical management of gout still faces numerous challenges. These include: a high recurrence rate, the inability to accurately predict gout recurrences, the unclear mechanism by which acute gout attacks progress to chronic inflammation, and the difficulty in effectively alleviating persistent joint damage and the formation of gout nodules. Addressing these issues requires a deeper understanding of the pathogenesis of gout.

[0004] However, most current research on the pathogenesis of gout focuses on the innate immune response and uric acid metabolism, particularly the pathway by which MSU crystals activate the NLR family pyrin domain-containing protein 3 (NLRP3) inflammasome. This process promotes the release of interleukin-1β (IL-1β) and the recruitment of neutrophils during acute gout attacks, triggering a dramatic inflammatory response. Although this mechanism is crucial for understanding acute gout attacks, a certain proportion of patients do not show significant MSU crystal deposition during relapses, and it does not fully explain the aforementioned unresolved clinical issues.

[0005] In order to solve the current problems in gout management, such as high recurrence rate and lack of effective prediction methods, there is an urgent need to develop reliable biomarkers and prediction models to achieve early identification of gout recurrence risk. Summary of the Invention

[0006] The purpose of this application is to overcome the shortcomings of the above-mentioned prior art and provide a gout recurrence prediction model and a method for constructing the same.

[0007] To achieve the above objectives, the technical solutions adopted in this application are:

[0008] This application provides a method for constructing a gout recurrence prediction model, comprising the following steps:

[0009] S1. Targeted proteomics was used to analyze the characteristics of immune-related molecules in plasma samples from healthy controls, gout intermittent patients, and gout acute patients in the training cohort.

[0010] S2. In the training cohort, identify differentially expressed proteins in each group through pairwise comparison. Then, combined with clinical indicator data, use mediation effect analysis to evaluate the mediating effects of multiple plasma protein molecules in the uric acid-mediated inflammation model.

[0011] S3. Analyze the plasma levels of differentially expressed immune-related molecules identified in the inter-gout and acute gout groups in the training cohort in the validation cohort. Then, perform Spearman correlation analysis on the training and validation cohorts to assess the relationship between the identified immune-related molecules and clinical inflammatory markers.

[0012] S4. Compare the proportion of patients in the training cohort who experience relapse within 12 weeks and 24 weeks after proteomic sample testing in the acute and remission phases of gout. Then, re-divide the training cohort into a relapse-prone group and a stable group based on whether the gout patients experience relapse within 12 or 24 weeks, and compare clinical indicator data and plasma protein levels. Perform feature screening using LASSO regression, and then statistically analyze the differences in the screened indicators between the relapse-prone group and the stable group. Draw receiver operating characteristic (ROC) curves for the screened indicators and calculate the area under the curve (AUC).

[0013] S5. In the validation cohort, verify the difference in recurrence rates between the acute and intermittent phases of gout at 12 and 24 weeks; use the logistic regression algorithm to construct the predictive efficacy of different combinations of the screened characteristic indicators, verify them in the training cohort, and calculate the accuracy and AUC value; use multiple machine learning algorithms to construct a gout recurrence prediction model based on characteristic indicators to verify the reliability of the logistic regression model.

[0014] As a reflection of systemic inflammatory status, plasma proteins have broad application prospects in disease prediction. Matrix metalloproteinase 1 (MMP1) is an important extracellular matrix-degrading enzyme that plays a key role in inflammatory responses, tissue remodeling, and immune cell migration. Studies have shown that MMP1 expression is upregulated in a variety of chronic inflammatory diseases and is closely associated with disease activity and relapse risk. Its expression changes in gout may reflect the underlying mechanism of relapse. Therefore, constructing a gout relapse prediction model based on plasma MMP1 levels will not only help identify high-risk patients and guide individualized treatment, but also improve long-term management outcomes and reduce the incidence of complications. The establishment of this model will promote the precise and forward-looking development of gout prevention and treatment, and has important clinical significance and application value.

[0015] In the technical solution of the present application, the present application uses clinical indicator data and the standardized protein expression (NPX) levels of immune-related molecules in plasma to construct a gout recurrence prediction model. The gout recurrence prediction model obtained in the present application has robust predictive ability, among which the MMP1 indicator achieves a very high AUC value and accuracy, as well as the lowest Brier score.

[0016] As a preferred embodiment of the method for constructing a gout recurrence prediction model described in the present application, compared with the healthy control group and the acute gout group, the immune-related molecules in the acute gout group include tumor necrosis factor-like cytokines, CXC motif chemokine ligand 1, CXC motif chemokine ligand 5, sirtuin 2, caspase 8 and sulfotransferase family 1A1, matrix metalloproteinase 1, calcium-binding protein 12, fibroblast growth factor 21, chemokine ligand 20 and at least one of interleukin 6.

[0017] As a preferred embodiment of the method for constructing a gout recurrence prediction model described in the present application, comparing the gout intermittent period group and the gout acute period group, the levels of matrix metalloproteinase 1, calcium binding protein 12, interleukin 6, CXC motif chemokine ligand 1 and CXC motif chemokine ligand 5 in the acute period gout group were higher than those in the gout intermittent period group.

[0018] In the technical solution of the present application, compared with the healthy control group, the plasma tumor necrosis factor-like cytokine (Oncostatin M, OSM) in the gout intermittent period group was significantly increased, while CXC motif chemokine ligand 1 (CXCL1), CXC motif chemokine ligand 5 (CXCL5), sirtuin 2 (SIRT2), caspase 8 (CASP8) and sulfotransferase family 1A1 (ST1A1) were significantly decreased. In the acute phase of gout with dampness syndrome, multiple immune molecules were significantly increased, including matrix metalloproteinase 1 (MMP1), calcium binding protein 12 (S100A12), fibroblast growth factor 21 (FGF21), chemokine ligand 20 (CCL20), OSM and interleukin 6 (IL-6), while ST1A1 was still in an inhibited state.

[0019] As a preferred embodiment of the method for constructing a gout recurrence prediction model described in the present application, in the validation cohort, the plasma levels of matrix metalloproteinase 1, calcium binding protein 12 and interleukin 6 in patients in the acute gout group were significantly higher than those in the intermittent gout group, which is consistent with the results of the training cohort.

[0020] As a preferred embodiment of the method for constructing a gout recurrence prediction model described in the present application, in step S2, the immune-related molecules screened are matrix metalloproteinase 1 (MMP1), calcium binding protein 12 (S100A12) and interleukin 6 (IL-6).

[0021] As a preferred embodiment of the method for constructing a gout recurrence prediction model described in the present application, in step S2, the clinical inflammatory markers include high-sensitivity C-reactive protein (hCRP) and erythrocyte sedimentation rate (ESR).

[0022] Experiments have shown that matrix metalloproteinase 1 (MMP1), calcium-binding protein 12 (S100A12) and interleukin 6 (IL-6) were elevated in plasma samples of the acute gout group, and were positively correlated with high-sensitivity C-reactive protein (hCRP) and erythrocyte sedimentation rate (ESR), indicating that these molecules may be involved in the inflammatory mechanism of gout.

[0023] As a preferred embodiment of the method for constructing a gout recurrence prediction model described in the present application, in step S2, a mediating effect model is used to confirm that the immune-related molecules finally obtained are matrix metalloproteinase 1, calcium-binding protein 12 and transforming growth factor α, which play key mediating effect molecules in the blood uric acid-mediated inflammation model.

[0024] To further investigate the role of plasma immune-related molecules in gout immune metabolic disorders, the study employed mediation effect models (MEMs). The results showed that SUA levels could regulate ESR through its effects on MMP1 and S100A12, with mediation effects of 52.39% and 60.23%, respectively, highlighting their roles in regulating chronic inflammation in gout.

[0025] At the same time, S100A12 also mediated the relationship between ESR and gout duration (years), with a mediation effect of 83.4%, suggesting that S100A12 may play a role in gout progression. In addition, SUA also increased hCRP through TGF-α, with a mediation effect of 72.4%.

[0026] Together, these results suggest a key role for the immune molecules MMP1 and S100A12 in the pathogenesis and progression of gout.

[0027] As a preferred embodiment of the method for constructing a gout recurrence prediction model described in the present application, in step S4, the clinical indicator data include body mass index and blood uric acid; the immune-related molecular indicator data include matrix metalloproteinase 1, CXC motif chemokine ligand 10 and CC motif chemokine ligand 28.

[0028] As a preferred embodiment of the method for constructing a gout recurrence prediction model described in the present application, the machine learning algorithm includes one of XGBoost, decision tree, logistic regression, K nearest neighbor and random forest.

[0029] This application first uses logistic regression in the training cohort, and constructs a prediction model for gout recurrence within 12 weeks with different combinations of the five indicators of BMI, MMP1, SUA, CXCL10 and CCL28, and then verifies it on the validation cohort, calculates the accuracy (Accuracy) and the AUC value. Among them, MMP has the best effect, and its logistic regression model has an accuracy of 0.6919 and an AUC value of 0.798 on the validation cohort. Subsequently, in the validation cohort, based on the variable MMP1, five different machine learning algorithms were used to construct a prediction model for gout recurrence within 12 weeks. All five models showed robust predictive ability. Among them, the logistic regression model performed better than other models, achieving the highest AUC value (0.915) and accuracy, as well as the lowest Brier score.

[0030] The present application provides a gout recurrence prediction model prepared by the above-mentioned gout recurrence prediction model construction method.

[0031] Compared with the prior art, this application has the following beneficial effects:

[0032] The present application provides a gout recurrence prediction model and a method for constructing the same. The present application uses clinical indicator data (body mass index) and the standardized protein expression (NPX) level of immune-related molecules (MMP1) in plasma to construct a gout recurrence prediction model. The gout recurrence prediction model obtained in the present application has robust predictive ability, achieving a very high AUC value and accuracy, as well as the lowest Brier score. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 The heat map and schematic diagram of differential protein expression of plasma proteins in healthy controls (HC), acute gout (GA-A), and intermittent gout (GA-R) in the training cohort are shown;

[0034] Figure 2 This is a diagram of the mediation effect model with immune-related molecules as mediating effect molecules in gout patients in the training cohort;

[0035] Figure 3 Schematic diagram of the expression levels of five plasma proteins in the acute phase (GA-A) and the intermittent phase (GA-R) of gout in the validation cohort;

[0036] Figure 4 Scatter plots showing the correlation between plasma protein expression levels and gout inflammation indicators in the training cohort (red) and validation cohort (green);

[0037] Figure 5 Comparison of recurrence rates between the acute phase (GA-A) and the intermittent phase (GA-R) of gout in the training cohort, LASSO regression screening of characteristic indicators for the recurrence group (Flare) and the stable group (Stable) within 12 weeks, intergroup difference analysis of characteristic indicators, and ROC result diagram;

[0038] Figure 6 To verify the comparison of the recurrence rates of gout in the acute phase (GA-A) and intermittent phase (GA-R) in the cohort, the accuracy and AUC values ​​of different models in the recurrence group (Flare) and stable group (Stable), the confusion matrix, and the result diagram of the prediction model based on MMP1. DETAILED DESCRIPTION

[0039] In order to better illustrate the purpose, technical solutions and advantages of this application, this application will be further described below with reference to the accompanying drawings and specific embodiments.

[0040] In the following examples, the experimental methods used are conventional methods unless otherwise specified, and the materials, reagents, etc. used are all commercially available unless otherwise specified, and the components and raw materials used in each parallel experiment are all the same.

[0041] Example 1

[0042] (1) Research materials:

[0043] 1. Sample source:

[0044] The study subjects included gout patients and healthy subjects. Gout patients were selected from a gouty arthritis cohort established in a clinical study, and age-matched healthy subjects were selected as a control group based on the research needs.

[0045] (2) Research methods:

[0046] 1. Experimental design:

[0047] This study, based on a gouty arthritis cohort established in a clinical setting, included age-matched patients in the acute phase of gout (GA-A, n=25) and those in the intermittent phase (GA-R, n=18) along with healthy controls (HC, n=9) as a training cohort. Furthermore, age-matched patients in the acute phase of gout (GA-A, n=46) and those in the intermittent phase (GA-R, n=40) served as a validation cohort. Targeted proteomics technology (O-Link) was used to quantitatively measure plasma protein levels in the training cohort (n=52) and validation cohort (n=86). Sample collection was approved by the Ethics Committee of Guangdong Provincial Hospital of Traditional Chinese Medicine (Approval Numbers BF2020-193-01 and BF2021-235-01).

[0048] 2. Research content:

[0049] The following information was collected from the patients: (1) general information: including name, gender, age, telephone number, medical card number, height, weight, body mass index (BMI), etc.; (2) relapse information: including relapse date and relapse frequency; (3) TCM dampness syndrome scale: a collection form for information on core symptoms of TCM dampness syndrome. Questions that the patients could not understand during the scale filling were completed under the guidance of rheumatologists, and tongue and pulse were determined by rheumatologists; (4) laboratory indicators: including erythrocyte sedimentation rate, C-reactive protein, three renal function tests, two liver function tests, three cardiac enzyme tests, fasting blood sugar, blood routine test, and urine routine test.

[0050] 3. Sample collection and processing:

[0051] Blood samples from patients with gouty arthritis (GA-A and GA-R) collected from the above clinical studies and healthy donors collected in this study were immediately shipped to the laboratory after collection.

[0052] The specific steps include:

[0053] Plasma separation: Blood samples were collected using EDTA-K2 anticoagulant vacuum blood collection tubes (BD Vacutainer) and centrifuged at 3000 rpm for 10 min. The upper plasma layer was separated, transferred to cryovials, and stored at −80°C. All samples were stored at the Biological Resource Center of Guangdong Provincial Hospital of Traditional Chinese Medicine according to standard operating procedures. All participants in this study provided written informed consent.

[0054] 4. Olink analysis and data processing:

[0055] A total of 138 plasma samples from healthy individuals and patients with gouty rheumatic syndrome were collected at enrollment, stored at -80°C, and thawed on ice before analysis. After thorough mixing, the samples were randomly assigned to 96-well PCR microplates. Proteomics measurements were performed using Olink Proximity Extension Assay (PEA) technology, which uses qPCR to simultaneously quantify a pre-designed protein panel in each sample. Specifically, this study used Target 96-well inflammation plate, proteomic detection, data normalization and quality control by BGI, China.

[0056] PEA technology utilizes antibodies labeled with specific oligonucleotides to bind to target proteins. Upon binding, the antibodies are drawn closer, generating a PCR sequence that allows for quantification of protein abundance via qPCR. Four internal controls are included in each sample to monitor assay performance and sample quality.

[0057] Quality control (QC) was performed in two steps: 1. The standard deviation of the internal control on each sample plate was assessed. Plates with a standard deviation less than 0.2 NPX were considered acceptable, and only data from these plates were included in the final analysis. 2. Sample quality was assessed by evaluating the deviation of each sample from the median of the internal control. Data from all samples that met QC criteria were included in the final dataset. Intra- and interassay coefficients of variation (CV) were calculated using a control sample (pooled plasma) included on each plate. These calculations were based on linear normalized protein expression (NPX) values, and the distribution of assays within the defined CV interval was displayed. Proteins with NPX values ​​below the limit of detection (in ≥20% of samples) were excluded from further analysis, resulting in a final dataset containing 75 proteins. NPX values ​​are expressed in log2 units and reflect relative protein abundance, where a difference of 1 NPX corresponds to a 2-fold change in protein expression.

[0058] 5. Receiver operating characteristic curve analysis (ROC analysis):

[0059] Receiver operating characteristic (ROC) curves were analyzed to assess the discriminatory ability of the variables. The multipleROC package in R was used to generate ROC curves for each independent variable. For each variable, the grouping variable (relapse-prone group or stable phase) was associated with the independent variable, and the ROC curve was calculated and plotted.

[0060] 6. Logistic regression algorithm development prediction model and evaluation:

[0061] In the training cohort, 31 different modeling combinations were constructed using the five variables (MMP1, SUA, CXCL10, CCL28, and BMI) as grouping variables, with relapse-prone group and stable group as grouping variables. Logistic regression using the glmnet package was used to construct these models. These 31 models were then validated in the validation cohort, and the accuracy and area under the curve (AUC) were calculated. The model constructed with MMP1 performed best, with the formula P(Group=1)=1 / (1+exp(-(-16.435+1.3547*MMP_1))).

[0062] 7. Machine Learning Model Development and Evaluation:

[0063] The analysis involved developing and comparing multiple machine learning models to classify the grouping variable (relapse-prone group vs. stable group) using the NPX data from MMP1. After the dataset was imported, the grouping variable was defined as the response variable, and the MMP1 variable was used as the predictor. Five machine learning models were considered: XGBoost, decision tree, logistic regression, K-nearest neighbor (KNN), and random forest. Each model was configured with the engine and classification mode using the tidymodels framework.

[0064] Create a workflow collection to combine the preprocessed dataset with the machine learning model. Use bootstrapping for 1000 iterations to ensure robustness of the evaluation results and save the predictions for further analysis. Evaluate model performance based on the area under the receiver operating characteristic curve (AUC). Results are sorted by AUC and visualized.

[0065] 8. Mediating effect analysis:

[0066] First, the R package mediation (version 4.5.0) was used to screen for exposure variables and mediating variables that were significantly associated with the specific outcome variable (P < 0.05 for the total effect). Then, a simple mediation effect analysis was performed using the mediate function in the R package mediation (version 4.5.0) to infer the causal role of the mediating variable in the change of the outcome variable.

[0067] 9. Correlation analysis:

[0068] The Spearman method was used to calculate the correlation coefficient r and its corresponding P value, and the ggplot2 package was used to generate the scatter plot. The points were colored and shaped according to the cohort, with custom colors added, and a linear trend line with a 95% confidence interval was added.

[0069] 10. Data entry and management:

[0070] A database was established using Excel, and data collected from patients during consultations and follow-up visits, as well as data extracted from the hospital information management system, were collated and entered. Information such as the patient's gender, age, consultation date, recurrence, TCM dampness syndrome score, and laboratory indicators were entered.

[0071] 11. Statistical analysis:

[0072] The measurement data that conform to the normal distribution are expressed as "mean ± standard deviation" (x ± s), and the measurement data that do not conform to the normal distribution are expressed as M (P 25 , P 75). Statistical analyses were performed using GraphPad Prism v9 or the R statistical software package. Two-way ANOVA was used for multivariate comparisons between multiple groups; one-way ANOVA was used for multivariate comparisons between two groups. Univariate comparisons between two groups were performed using the independent sample t-test for normally distributed data and the nonparametric Mann-Whitney U test for data that did not. The Kruskal-Wallis H test was used for multi-group comparisons. Specific statistical methods used for each data set are detailed in the corresponding figure and table legends. Correlation analyses were performed using Spearman correlation analysis. P < 0.05 was considered statistically significant.

[0073] (2) Results:

[0074] 1. Expression of immune-related molecules in gout patients, mediating effect model and their relationship with clinical indicators:

[0075] 1) MMP1, S100A12, IL-6, CXCL1, and CXCL5 were elevated in GA-A plasma in the training cohort.

[0076] The study used targeted proteomics technology to analyze the characteristics of immune-related molecules in plasma samples of the training cohort (healthy control group (HC), gout inter-stage group (GA-R) and gout acute stage group (GA-A)).

[0077] The results are as follows Figure 1 shown.

[0078] in, Figure 1 Figure A is a heat map showing the expression levels of all detected plasma immune-related molecules in HC, GA-A, and GA-R in the training cohort.

[0079] Figure 1 Middle B shows a forest plot showing the mean difference in NPX between the GA-R and HC groups in the training cohort. Statistically significant markers are annotated with adjusted P values; proteins significantly upregulated in GA-R are shown to the right of the mean difference axis, while downregulated proteins are to the left.

[0080] Figure 1 Middle C shows a forest plot showing the mean difference in NPX between the GA-A and HC groups in the training cohort. Statistically significant markers are annotated with adjusted P values; proteins significantly upregulated in GA-A are shown to the right of the mean difference axis, while downregulated proteins are to the left.

[0081] Figure 1Middle D shows a forest plot showing the mean difference in NPX between the GA-A and GA-R groups in the training cohort. Statistically significant markers are annotated with adjusted P values; proteins significantly upregulated in GA-A are shown to the right of the mean difference axis, while downregulated proteins are to the left.

[0082] The conclusion is:

[0083] Figure 1 The heat map in Figure A reveals unique serum immune-related molecular profiles between different groups. Compared with HC, GA-R plasma levels of tumor necrosis factor-like cytokines (Oncostatin M, OSM) were significantly increased, while CXC motif chemokine ligand 1 (CXCL1), CXC motif chemokine ligand 5 (CXCL5), sirtuin 2 (SIRT2), caspase 8 (CASP8), and sulfotransferase family 1A1 (ST1A1) were significantly decreased ( Figure 1 Compared with HC, GA-A plasma showed significant increases in multiple immune molecules, including matrix metalloproteinase 1 (MMP1), calcium binding protein 12 (S100A12), fibroblast growth factor 21 (FGF21), chemokine ligand 20 (CCL20), OSM, and interleukin 6 (IL-6), while ST1A1 remained in an inhibited state ( Figure 1 Comparison of GA-A and GA-R revealed that the levels of MMP1, S100A12, IL-6, CXCL1, and CXCL5 were higher during the acute inflammatory phase ( Figure 1 Middle D).

[0084] 2) MMP1 and S100A12 in the plasma of gout patients in the training cohort can mediate the blood uric acid-mediated inflammation model.

[0085] Figure 2 Figure A shows the mediation model evaluating the indirect effect of SUA on ESR via MMP1. Each triangular diagram depicts the path between the predictor variable (X), mediator variable (M), and outcome variable (Y), and displays the percentage mediation (ratio of the indirect effect to the total effect) and corresponding P-value for the mediation model (purple arrow path) and the inverse mediation model (green arrow path).

[0086] Figure 2 Middle B is a mediation model evaluating the indirect effect of SUA on ESR through S100A12.

[0087] Figure 2 Middle C is a mediation model that evaluates the indirect effect of disease duration (Duration / year) on ESR through S100A12.

[0088] Figure 2Figure D is a mediation model evaluating the indirect effect of SUA on hCRP through TGFα.

[0089] The conclusion is:

[0090] To evaluate whether immune-related molecules are involved in the inflammatory state mediated by serum uric acid (SUA), a mediation effect model (MEM) was used. The results showed that SUA levels can regulate ESR through its effects on MMP1 and S100A12, with mediation effects of 52.39% and 60.23%, respectively. Figure 2 China A and Figure 2 B), emphasizing their role in regulating chronic inflammation in gout. At the same time, S100A12 also mediated the relationship between ESR and gout duration (years), with a mediating effect of 83.4% ( Figure 2 C), indicating that S100A12 may play a role in the progression of gout. In addition, SUA also increased hCRP through TGF-α, with a mediating effect of 72.4% ( Figure 2 Middle D).

[0091] 3) Elevated MMP1, S100A12, and IL-6 in GA-A plasma in the validation cohort

[0092] Figure 3 Figure A is a scatter plot showing the NPX levels of MMP1, S100A12, IL-6, CXCL1, and CXCL5 in the validation cohorts (GA-R (n=40) and GA-A (n=46)); P values ​​were obtained by unpaired t-test.

[0093] The conclusion is:

[0094] In the validation cohort, MMP1, S100A12, and IL-6 in the GA-A group were significantly higher than those in the GA-R group, and the differences were statistically significant, which was consistent with the results in the training group.

[0095] 4) Correlation analysis between plasma immune-related molecules and clinical indicators in gout patients:

[0096] Figure 4 Figure A shows the correlation analysis between NPX levels of immune molecules (MMP1, S100A12, IL6) and hCRP in test cohort 1 (circles) and validation cohort 2 (triangles). For visualization, regression lines are drawn and the shaded area represents the 95% confidence interval.

[0097] Figure 4Figure B shows the correlation analysis between NPX levels of immune molecules (MMP1, S100A12, IL6) and ESR in test cohort 1 (circles) and validation cohort 2 (triangles). For visualization, regression lines are drawn and the shaded area represents the 95% confidence interval.

[0098] The conclusion is:

[0099] Spearman correlation analysis was used to evaluate the relationship between MMP1, S100A12, and IL-6 and clinical inflammatory markers high-sensitivity C-reactive protein (hCRP) and erythrocyte sedimentation rate (ESR). In both cohorts, MMP1, S100A12, and IL-6 were significantly positively correlated with hCRP and ESR ( Figure 4 China A and Figure 4 Middle B), suggesting that these molecules may be involved in the inflammatory mechanism of gout.

[0100] Together, these results suggest a key role for the immune molecules MMP1 and S100A12 in the pathogenesis and progression of gout.

[0101] 2. Construction of a prediction model for gout recurrence:

[0102] 1) Screening of characteristic variables for predicting recurrence in the training cohort:

[0103] The study attempted to use clinical indicator data and NPX levels of plasma immune molecules to construct a predictive model for gout recurrence.

[0104] To further investigate the impact of different disease stages on relapse in gout patients, we calculated the proportion of relapses within 12 and 24 weeks after baseline (the time point when plasma was collected for targeted proteomics) in the training cohort. We then regrouped patients based on whether they relapsed within 12 weeks; those who relapsed were designated the flare group (n=19) and those who did not relapse were designated the stable group (n=21).

[0105] In order to screen the characteristic indicators with modeling potential, first, eight non-zero variables were identified by Lasso regression: BMI, MMP1, CCL28, CXCL10, CD6, SUA, IFN-gamma and MCP_3 (see Figure 5 Then, the inter-group expression differences of these eight indicators were analyzed, and the results showed that five indicators, including BMI, MMP1, CCL28, CXCL10, and SUA, were increased in the relapse-prone group in the training cohort ( Figure 5 ROC curves were generated for BMI, MMP1, SUA, CCL28, and CXCL10, with AUC values ​​of 0.798, 0.787, 0.704, 0.704, and 0.702, respectively ( Figure 5 Middle D).

[0106] The results are as follows Figure 5 shown.

[0107] Figure 5 Figure A: Bar chart showing the number of patients with gout recurrence (purple) and without recurrence (gray) in the GA-A and GA-R groups in the training cohort during 12 weeks (12W) and 24 weeks (24W) of follow-up.

[0108] Figure 5 Middle B is: LASSO regression model effect diagram of variables that can distinguish the relapse-prone group from the stable group.

[0109] Figure 5 Middle C: The box plot shows the differences in 8 non-zero variables screened by LASSO regression between the gout relapse-prone group and the stable group within 12 weeks; statistical comparison was performed using the Mann-Whitney U test.

[0110] Figure 5 D in the middle is: ROC curve of 5 variables, including AUC, 95% confidence interval, optimal cutoff value and P value.

[0111] 2) Development and validation of a gout recurrence prediction model:

[0112] To validate the results of the training set, the study first calculated the recurrence rates of acute gout and intermittent gout in the validation cohort. Similarly, it was found that patients with acute gout were more likely to relapse in the short term (12 weeks and 24 weeks), with recurrence rates of 65.63% and 76.67%, respectively, higher than those in the stable phase (40.63% and 51.72%). Figure 6 In order to develop a prediction model for gout recurrence, the five indicators (BMI, MMP1, SUA, CCL28, and CXCL10) screened in the training cohort were combined in different ways to form a total of 31 combination variables. Then, 31 corresponding models were constructed using the logistic regression algorithm. These 31 models were validated using the validation cohort, and the accuracy and AUC values ​​of the models on the validation cohort were calculated ( Figure 6 In the middle B), the gout recurrence prediction model with MMP1 as the variable was the best, with an accuracy of 67.19% and an AUC value of 0.798 ( Figure 6 Finally, five machine learning algorithms were used to further verify the potential of using MMP1 as a variable to construct a gout recurrence prediction model in the validation cohort. Among them, the logistic regression model performed better than other models, achieving the highest AUC value (0.915) and accuracy, as well as the lowest Brier score ( Figure 6 Middle D).

[0113] The results are as follows Figure 6 shown.

[0114] Figure 6 Figure A: The bar chart shows the number of patients with gout recurrence (purple) and no recurrence (gray) in the GA-A and GA-R groups in the validation cohort during 12 weeks (12W) and 24 weeks (24W) of follow-up.

[0115] Figure 6 Middle B: The bar chart shows the effectiveness of the gout recurrence prediction model for 31 different combinations, including accuracy (dark blue) and AUC value (yellow).

[0116] Figure 6 C in the middle: The confusion matrix and ROC curve show the effectiveness of the gout recurrence prediction model with MMP1 as the variable.

[0117] Figure 6 Figure D: Bar chart comparing the accuracy, Brier score, and AUC of five machine learning models (boosting tree, decision tree, logistic regression, nearest neighbor, and random forest) when applied in a validation cohort (n = 64) using MMP1 as a variable for predicting gout recurrence within 12 weeks. Error bars represent the standard deviation of resampling iterations.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit the scope of protection of the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application may be modified or replaced by equivalents without departing from the essence and scope of the technical solutions of the present application.

Claims

1. A method for constructing a gout recurrence prediction model, characterized in that: The following steps are involved: S1. Targeted proteomics was used to analyze the characteristics of immune-related molecules in plasma samples from healthy controls, gout intermittent patients, and gout acute patients in the training cohort. S2. In the training cohort, identify differentially expressed proteins in each group through pairwise comparison. Then, combined with clinical indicator data, use mediation effect analysis to evaluate the mediating effects of multiple plasma protein molecules in the uric acid-mediated inflammation model. S3. Analyze the plasma levels of differentially expressed immune-related molecules identified in the inter-gout and acute gout groups in the training cohort in the validation cohort. Then, perform Spearman correlation analysis on the training and validation cohorts to assess the relationship between the identified immune-related molecules and clinical inflammatory markers. S4. Compare the proportion of patients in the training cohort who experience relapse within 12 weeks and 24 weeks after proteomic sample testing in the acute and intermittent phases of gout. Then, re-divide the training cohort into a relapse-prone group and a stable group based on whether the gout patients experience relapse within 12 or 24 weeks, and compare clinical indicator data and plasma protein levels. Perform feature screening using LASSO regression, and then statistically analyze the differences in the screened indicators between the relapse-prone group and the stable group. Draw receiver operating characteristic (ROC) curves for the screened indicators and calculate the area under the curve (AUC). S5. In the validation cohort, verify the difference in recurrence rates between the acute and intermittent phases of gout at 12 and 24 weeks; use the logistic regression algorithm to construct the predictive efficacy of different combinations of the screened characteristic indicators, verify them in the training cohort, and calculate the accuracy and AUC value; use multiple machine learning algorithms to construct a gout recurrence prediction model based on characteristic indicators to verify the reliability of the logistic regression model.

2. The method for constructing a gout recurrence prediction model according to claim 1, wherein: Compared with the healthy control group and the acute gout group, the immune-related molecules in the acute gout group included tumor necrosis factor-like cytokine, CXC motif chemokine ligand 1, CXC motif chemokine ligand 5, sirtuin 2, caspase 8 and sulfotransferase family 1A1, matrix metalloproteinase 1, calcium-binding protein 12, fibroblast growth factor 21, chemokine ligand 20 and at least one of interleukin 6.

3. The method for constructing a gout recurrence prediction model according to claim 2, wherein: Comparing the gout intermittent group with the gout acute group, the levels of matrix metalloproteinase-1, calcium binding protein-12, interleukin-6, CXC motif chemokine ligand 1, and CXC motif chemokine ligand 5 in the gout acute group were higher than those in the gout intermittent group.

4. The method for constructing a gout recurrence prediction model according to claim 3, wherein: In the validation cohort, the plasma levels of matrix metalloproteinase-1, calcium-binding protein-12, and interleukin-6 in patients in the acute gout group were significantly higher than those in the intermittent gout group, which was consistent with the results of the training cohort.

5. The method for constructing a gout recurrence prediction model according to claim 1, wherein: In step S3, the immune-related molecules screened are matrix metalloproteinase 1, calcium binding protein 12 and interleukin 6.

6. The method for constructing a gout recurrence prediction model according to claim 1, wherein: In step S2, clinical inflammatory markers include high-sensitivity C-reactive protein and erythrocyte sedimentation rate.

7. The method for constructing a gout recurrence prediction model according to claim 1, wherein: In step S2, a mediating effect model is used to confirm that the immune-related molecules matrix metalloproteinase-1, calcium binding protein 12 and transforming growth factor α play the role of key mediating effect molecules in the blood uric acid-mediated inflammation model.

8. The method for constructing a gout recurrence prediction model according to claim 1, wherein: In step S4, the clinical indicator data include body mass index and blood uric acid; the immune-related molecular indicator data include matrix metalloproteinase 1, CXC motif chemokine ligand 10, and CC motif chemokine ligand 28.

9. The method for constructing a gout recurrence prediction model according to claim 1, wherein: The machine learning algorithm includes one of XGBoost, decision tree, logistic regression, K nearest neighbor and random forest.

10. A gout recurrence prediction model prepared by the method for constructing a gout recurrence prediction model according to any one of claims 1 to 9.

Images