Establishment method and evaluation method of post-stroke inflammatory injury prediction model

Abstract

The invention discloses an establishment method and an evaluation method of a post-stroke inflammation injury prediction model, and belongs to the technical field of post-stroke inflammation injury prediction, and the establishment method of the post-stroke inflammation injury prediction model comprises the following steps: 1, collecting clinical index data of an AIS group and a health control group; 2, after the AIS group and the healthy control group are subjected to tendency score matching according to gender and age, a matched AIS group with the same number of people as that of the healthy control group is obtained, and matching features of the healthy control group and the matched AIS group are obtained; step 3, difference and correlation analysis of matching characteristics between the healthy control group and the matched AIS group; 4, carrying out single factor analysis on fibrinogen level influence indexes in the AIS group; and step 5, inputting the screened confounding factors into a full-automatic machine learning model for model training and verification to obtain a post-stroke inflammation injury prediction Logistic regression model. The regression model established through the steps is good in sensitivity and high in accuracy when being used for predicting the post-stroke inflammatory injury.

Description

Technical Field

[0001] This invention belongs to the field of post-stroke inflammatory injury prediction technology, specifically involving the establishment and evaluation methods of post-stroke inflammatory injury prediction models. Background Technology

[0002] Following an acute ischemic stroke (AIS), a cascade of inflammatory storms occurs in the brain parenchyma. This process generates complex inflammation-related proteins, leading to the infiltration and activation of various inflammatory cells. These inflammatory components and infiltrating cells primarily alter the brain tissue microenvironment, which cannot be detected and analyzed using imaging techniques in the short term. However, the expression levels of these protein molecules may change immediately in the peripheral blood or cerebrospinal fluid, dynamically reflecting the progression of the inflammatory response in the central nervous system. Detecting these inflammatory markers or monitoring changes in the proportion of inflammatory cells could be a rapid and practical method for monitoring the level of inflammatory response in the brain parenchyma after infarction. Therefore, many studies currently focus on how to predict stroke prognosis and control the inflammatory response through changes in blood inflammatory markers.

[0003] Previous studies have confirmed a positive correlation between the expression levels of various inflammatory factors and post-infarction neurological function impairment. For example, many clinical studies on AIS patients have found that using the National Institutes of Health Stroke Scale (NIHSS) to evaluate neurological function, patients with high NIHSS scores showed significantly elevated levels of inflammatory markers such as IL-6, C-reactive protein (CRP), and TNF-α in peripheral blood or cerebrospinal fluid; these inflammatory markers were also associated with poor prognosis. However, earlier studies found no correlation between inflammatory factors and lesion size or prognosis; additionally, some cytokines, such as IL-6, play a dual role in inflammation and neuroprotection at different stages of the inflammatory response following acute ischemia. These clinical studies varied in the number of cases, disease duration, sampling time, and follow-up time, and the conclusions reflected significant heterogeneity in the expression patterns and functional effects of cytokines among stroke patients. Therefore, further in-depth research is needed to understand how various inflammatory factors specifically act as markers of inflammatory responses.

[0004] In recent years, more and more perspectives have emerged suggesting that multiple factors interact during the inflammatory response, amplifying intercellular signals. Analyzing only one factor may encounter problems such as significant heterogeneity and a limited analytical perspective. Therefore, many studies have focused on jointly analyzing multiple inflammatory markers or cell counts to construct analytical models for assessing the severity of inflammation and predicting prognosis after ischemic stroke. However, most of these attempts remain limited to small-sample population studies, and the exploration and identification of post-stroke inflammatory biomarkers still face many challenges.

[0005] Previous animal studies have confirmed that fibrinogen deposits in the infarct area after acute ischemia in a middle cerebral artery occlusion (MCAO) animal model, activating neutrophils and exacerbating post-infarction inflammation and secondary brain injury. However, these damaging responses occur in the central nervous system, making timely clinical monitoring difficult. We note that fibrinogen is also a typical indicator of inflammation in peripheral blood. Peripheral blood fibrinogen levels have been confirmed by numerous studies to be associated with the pathogenesis and prognosis of acute myocardial infarction (AIS). Fibrinogen levels gradually increase within 24 hours of acute onset of AIS, and high fibrinogen levels in peripheral blood of AIS patients at admission are associated with poor prognosis. These phenomena overlap to some extent with the central nervous system inflammatory responses we observed in the MCAO model. In conclusion, changes in peripheral blood fibrinogen levels in AIS patients can reflect, to some extent, the dynamic changes in central nervous system inflammation and the severity of central inflammatory damage, but further analysis and verification are needed. Summary of the Invention

[0006] To address the technical problems mentioned above, this invention proposes a method for establishing and evaluating a predictive model for post-stroke inflammatory damage. While collecting peripheral blood fibrinogen data from AIS patients, the invention incorporates analysis of neuron-specific enolase (NSE) expression levels. It reveals a correlation between changes in fibrinogen levels in the peripheral circulation during the acute phase of AIS and brain tissue inflammatory damage. A logistic regression model for predicting post-stroke inflammatory damage based on fibrinogen levels is established to assess the extent of post-stroke inflammatory damage, demonstrating high accuracy and stability.

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

[0008] One object of the present invention is a method for establishing a prediction model of post-stroke inflammatory damage, characterized in that the method includes the following steps:

[0009] Step 1: Collect clinical indicator data from the AIS group and the healthy control group;

[0010] Step 2: After matching the AIS group and the healthy control group according to their propensity scores based on gender and age, a matched AIS group with the same number of members as the healthy control group is obtained, and the matching characteristics between the healthy control group and the matched AIS group are obtained.

[0011] Step 3: Analysis of differences and correlations in matching characteristics between the healthy control group and the matched AIS group;

[0012] Step 4: Based on the analysis in Step 3, conduct a univariate analysis of the influencing factors of fibrinogen levels in the AIS group;

[0013] Step 5: Input the confounding factors identified in Step 4 into the fully automated machine learning model for model training and validation to obtain a Logistic regression model for predicting post-stroke inflammatory damage based on fibrinogen levels.

[0014] Furthermore, the clinical data collected in step 1 include gender, age, past medical history, complete blood count, serum biochemical indicators, blood lipid indicators, fibrinogen level, D-dimer, partial thromboplastin time, prothrombin time, NSE, tumor markers, and treatment received before enrollment.

[0015] Furthermore, past medical history includes hypertension, coronary heart disease, diabetes, and atrial fibrillation; routine blood tests include red blood cell count, white blood cell count, lymphocyte count, neutrophil count, and platelet count.

[0016] Furthermore, the matching features obtained in step 2 include gender, age, white blood cell count, neutrophil count, fibrinogen level, NSE, hypertension, diabetes, coronary heart disease, and atrial fibrillation.

[0017] Furthermore, step 3 specifically includes the following steps:

[0018] Step 301: Analysis of differential expression levels of NSE, neutrophils, and fibrinogen between the healthy control group and the matched AIS group;

[0019] Step 302: ROC curve analysis of NSE and fibrinogen between the healthy control group and the matched AIS group;

[0020] Step 303: Correlation analysis of fibrinogen and NSE between the healthy control group and the matched AIS group.

[0021] Furthermore, before conducting univariate analysis of fibrinogen levels in AIS patients, it is necessary to analyze the correlation between fibrinogen and NSE expression levels in AIS patients and determine the dependence of fibrinogen and NSE expression levels in AIS patients.

[0022] Furthermore, the confounding factors identified by the univariate analysis in step 4 included sex, age, neutrophils, monocytes, platelets, serum albumin, DD dimer, and thrombin time.

[0023] Furthermore, following the univariate analysis in step 4, a curve fitting analysis was performed to determine the relationship between fibrinogen and central nervous system cell damage markers in AIS patients, as detailed below:

[0024] Confounding factors screened by univariate analysis were incorporated into the smooth curve fitting analysis of fibrinogen and NSE to determine that the degree of central nervous cell damage in AIS patients increased with increasing fibrinogen levels.

[0025] Furthermore, the logistic regression model established in step 5 for predicting post-stroke inflammatory damage based on fibrinogen levels is as follows: y = -7.00338 + 5.62873 × (1 / fibrinogen) 2 )-27.20105×[1 / fibrinogen 2 [×log(fibrinogen)] + 0.76015 × (male = 1) + 0.04274 × age + 0.07542 × monocyte count + 0.00343 × platelet count + 0.04120 × albumin + 0.11142 × thrombin time.

[0026] Another object of the present invention is to provide a method for assessing the degree of post-stroke inflammatory damage using the established post-stroke inflammatory damage prediction model, comprising the following steps:

[0027] (1) Collect the patient's fibrinogen level, monocyte count, platelet count, albumin count and thrombin time;

[0028] (2) The collected fibrinogen level, monocyte count, platelet count, albumin count, thrombin time, and the patient's age and gender were input into the Logistic regression model to calculate the predicted value of post-stroke inflammatory damage.

[0029] Compared with the prior art, the present invention has the following advantages:

[0030] The logistic regression model established in this invention for predicting post-stroke inflammatory damage based on fibrinogen levels is: y = y = -7.00338 + 5.62873 × (1 / fibrinogen) 2 )-27.20105×[1 / fibrinogen 2[×log(fibrinogen)] + 0.76015 × (male = 1) + 0.04274 × age + 0.07542 × monocyte count + 0.00343 × platelet count + 0.04120 × albumin + 0.11142 × thrombin time. The above regression model has good sensitivity and high accuracy in predicting post-stroke inflammatory damage. Attached Figure Description

[0031] Figure 1 The research subjects are incorporated into the flowchart.

[0032] Figure 2 The differential expression levels of NSE and fibrinogen in the healthy control group and the matched AIS group; (A) Comparison of NSE levels between the healthy control group and the matched AIS group; (B) Comparison of fibrinogen expression levels between the healthy control group and the matched AIS group.

[0033] Figure 3 ROC curve analysis of NSE and fibrinogen between the healthy control group and the matched AIS group; (A) ROC curve analysis of NSE level between the healthy control group and the matched AIS group; (B) ROC curve analysis of fibrinogen expression level between the healthy control group and the matched AIS group.

[0034] Figure 4 Correlation analysis of fibrinogen and NSE in healthy controls and matched AIS groups;

[0035] Figure 5 Correlation analysis of fibrinogen and NSE in AIS patients.

[0036] Figure 6 Correlation analysis of fibrinogen and NSE in AIS patients; (A) After adjusting for confounding factors, fibrinogen and NSE showed a linear relationship in AIS patients; (B) Stratified analysis according to NSE expression level showed that high fibrinogen level was still linearly related to high NSE level.

[0037] Figure 7 Logistic regression model for predicting post-stroke inflammatory damage based on fibrinogen levels.

[0038] Figure 8 Overall evaluation of a fully automated machine learning prediction model for predicting post-stroke inflammatory damage based on fibrinogen levels.

[0039] Figure 9 Evaluation of different algorithms for a fully automated machine learning prediction model for predicting post-stroke inflammatory damage based on fibrinogen levels. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0041] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0042] 1. Research Subjects and Methods

[0043] 1.1 Patient Enrollment Ethics

[0044] In accordance with the Declaration of Helsinki, this study was approved by the Ethics Committee of Renji Hospital Baoshan Branch, affiliated with Shanghai Jiao Tong University School of Medicine, China (Ethics Approval No. 2022-KSSC-01). All study participants or their immediate family members provided informed consent before sample collection.

[0045] 1.2 Research Design

[0046] Patients with AIS who were hospitalized at the Baoshan Branch of Renji Hospital affiliated with Shanghai Jiao Tong University School of Medicine from January 1, 2021 to August 31, 2022 were continuously included, and their data were entered into the hospital's stroke registry database.

[0047] 1.3 Research Subjects

[0048] Patients with AIS are diagnosed according to World Health Organization Standard 16.

[0049] The patient inclusion criteria are as follows:

[0050] (1) AIS develops within 24 hours;

[0051] (2) It can assess clinical symptoms and signs;

[0052] (3) Confirmed by cranial imaging computed tomography (CT) or magnetic resonance imaging (MRI);

[0053] (4) 40 years old or above.

[0054] The exclusion criteria are as follows:

[0055] (1) Transient ischemic attack;

[0056] (2) Cerebral hemorrhage;

[0057] (3) After malignant tumors and splenectomy;

[0058] (4) Primary thrombocytopenic purpura, megaloblastic anemia, post-splenectomy, leukemia, megaloblastic syndrome and aplastic anemia;

[0059] (5) Valvular heart disease and acute myocardial infarction;

[0060] (6) Clinical and laboratory data cannot be used for analysis, including unintegrated patient data.

[0061] Healthy control subjects were recruited and matched for age and sex using propensity score matching. The exclusion criteria for healthy control subjects who underwent health screening are as follows:

[0062] (1) An infection or inflammatory event has occurred within the past 4 weeks;

[0063] (2) Abnormal serum lipids, blood glucose, liver function and kidney function;

[0064] (3) A history of ischemic stroke, myocardial infarction, thrombophlebitis, and related medical history that may affect changes in neutrophil and fibrinogen levels.

[0065] 1.4 Clinical Data Collection and Laboratory Testing Methods

[0066] Clinical data collection:

[0067] Upon admission, we interviewed the patient and their family to collect detailed baseline data on demographic characteristics, medical history, and medications used prior to admission.

[0068] Medical history includes: hypertension, coronary heart disease (CHD), diabetes, and atrial fibrillation;

[0069] The medications used before admission included: antidiabetic drugs, antihypertensive drugs, lipid-lowering drugs, anticoagulants, and antiplatelet drugs.

[0070] Laboratory testing methods:

[0071] Venous blood samples were collected upon admission and prior to treatment, including intravenous administration of recombinant tissue plasminogen activator (r-tPA) and any angioplasty performed in the emergency room. Blood samples were collected in vacuum tubes containing EDTA to assess fibrinogen levels, which were measured using a commercially available fibrinogen kit (semi-automated coagulation analyzer) purchased from Biotechnology Co., Ltd. (Shanghai, China). The intra-assay and inter-assay coefficients of variation were 2.3% and 5.34%, respectively, while the detection limits for fibrinogen were 0.39 to 25.0 g / L, and the normal reference range for fibrinogen was 2 g / L to 4 g / L.

[0072] NSE levels were assessed using a commercially available quantitative sandwich enzyme-linked immunosorbent assay (ELISA) kit purchased from the R&D system (Shanghai, China). Intra-assay and inter-assay coefficients of variation were b = 3% and b = 7%, respectively. The limit of detection was 0.229 ng / mL, the detection range for NSE was 0.625 to 40 ng / mL, and the normal laboratory NSE value was <16.3 mg / L.

[0073] Fasting blood samples are collected via venipuncture, and the following indicators are routinely tested:

[0074] (1) Blood tests: red blood cell (RBC) count, white blood cell count, lymphocyte count, neutrophil count and platelet count;

[0075] (2) Serum biochemical indicators: alanine aminotransferase, total bilirubin, fasting blood glucose, creatine, uric acid, urea, homocysteine, glycated hemoglobin and erythrocyte sedimentation rate;

[0076] (3) Blood lipids: triglyceride, low-density lipoprotein cholesterol (LDL-C) and high-density lipoprotein cholesterol (HDL-C) levels;

[0077] (4) Coagulation index: fibrinogen, D-dimer, partial thromboplastin time, prothrombin time;

[0078] (5) Special blood marker: NSE.

[0079] All measurements were performed by laboratory technicians without knowing the patient's detailed information.

[0080] Grouped into 1.5

[0081] All enrolled patients were grouped according to two criteria.

[0082] (1) First, in the baseline characteristic analysis, patients were grouped according to their fibrinogen levels. When the fibrinogen value was greater than 4 g / L, it was judged as high fibrinogen. In addition, according to the statistical trigonometric method, patients were divided into T1 (low), T2 (medium) and T3 (high) groups.

[0083] (2) Second, based on the normal reference values ​​of laboratory indicators in the stratified analysis, including fasting blood glucose, LDL-C, and red blood cells, the data were grouped. The data of ALT, BUN, creatinine, lymphocyte count, and platelet count were divided into low and high groups respectively using a dichotomy.

[0084] 1.6 Propensity Score Matching

[0085] Propensity score matching was performed based on baseline covariates to reduce bias between AIS and healthy controls and to control for potential confounding factors. Therefore, our study used a nearest neighbor matching algorithm adjusted for sex and age, with a tolerable caliper width of 0.01, and 1:1 individual matching. We compared neutrophil and fibrinogen levels between the AIS group and the healthy controls and described the correlations between neutrophils and fibrinogen in the AIS group and the healthy controls after propensity score matching.

[0086] 1.7 Statistical Analysis Methods

[0087] Baseline characteristics of patients included in the study were described according to fibrinogen levels. Statistical analysis was performed using SPSS 21.0 software. Count data were expressed as percentages (%), and comparisons between groups were performed using the chi-square test or Fisher's exact test. Normally distributed continuous data were expressed as mean ± standard deviation, and comparisons between two groups were performed using the independent samples t-test. Comparisons among multiple groups were performed using ANOVA, and pairwise comparisons among multiple groups were performed using the one-way ANOVA Dunnett T3 test. Non-normally distributed continuous data were expressed as median and interquartile range, and comparisons between two groups were performed using the Mann-Whitney U test. Comparisons among multiple groups were performed using the Kruskal-Wallist rank-sum test, and pairwise comparisons among multiple groups were performed using the one-way ANOVA Dunnett T3 test. Correlation between fibrinogen, neutrophils, and NSE was analyzed using Pearson correlation analysis (for continuous variables).

[0088] Statistical analysis was performed on the differences in fibrinogen between the AIS group and the healthy control group. Receiver operating characteristic (ROC) curve analysis was used to validate these differences, and the area under the ROC curve (AUC) of the 95% confidence interval (CI) was calculated. An AUC of 1.0 indicated complete discriminative significance, while an AUC < 0.5 indicated no discriminative significance. The relationship between fibrinogen and NSE was assessed using linear curve fitting analysis (generalized additive model). Baseline variables showing a univariate relationship with high fibrinogen levels were included as confounding factors in the curve fitting analysis model, and stratified analysis was performed by gender. The statistical significance level was set at two-tailed p < 0.05.

[0089] 1.8 Steps for establishing a prediction model:

[0090] This module uses the Rpackage h2o automl (Automate Machine Learning) feature.

[0091] Usage requirements:

[0092] 1. The outcome variable is binary (coded as 0 / 1), multi-category, or continuous.

[0093] 2. Independent variables can be binary, multi-category, or continuous. Input variables must be numeric; if they are character-based, they must be re-encoded into numbers (using Easy's categorical variable value recombination function).

[0094] Main parameter settings:

[0095] Internal validation set: The purpose of validation is to prevent overfitting.

[0096] 1. Use 5-group internal cross-validation (default option 0);

[0097] 2. Alternatively, randomly select a group as the validation set, and select or enter the percentage, such as 25% or 15%. Manual input can be done by entering only the number (without the "%" sign, such as 20).

[0098] 3. Alternatively, define a set as the validation set, such as manually inputting POP=1. (Inputting POP==1 has the same effect as POP=1).

[0099] Model testing set: The purpose of testing is to perform external validation of the model.

[0100] 1. Randomly select a group to serve as the validation set. Select or enter the percentage, such as 25% or 15%. Manual input can be done by entering only the number (without the "%" sign, such as 20).

[0101] 2. Alternatively, define a set as the validation set, such as manually inputting POP=2. (Inputting POP==2 has the same effect as POP=2).

[0102] Algorithm selection: By default, various applicable algorithms are selected based on the type of the result variable. Alternatively, you can select algorithms such as deep learning (fully connected deep neural network), random forest including DRF (distributed random forest) and XRT (extremely random trees), GBM (gradient boosting machine), and GLM (generalized linear model).

[0103] Automatic model comparison and selection criteria:

[0104] For the outcome variables of the two classifications, compare "AUC", "AUCPR", and "logloss".

[0105] "mean_per_class_error","RMSE","MSE". The model with the highest AUC is selected as the best model by default.

[0106] For continuous outcome variables, compare "mean_residual_deviance", "RMSE", "MSE", "MAE", and "RMSLE". The model with the lowest mean_residual_deviance is selected as the best model by default.

[0107] For multi-class outcome variables, compare "mean_per_class_error", "logloss", "RMSE", and "MSE". By default, the model with the lowest mean_per_class_error is selected as the best model.

[0108] Output result:

[0109] Automated machine learning automatically runs multiple models, such as 20 models selected from various applicable algorithms, or 10 models selected from a single algorithm. The best model is automatically selected through comparison. The performance and prediction results of the selected best model are then analyzed, and the output includes:

[0110] Models comparison

[0111] The following results are based on the automatically selected optimal model:

[0112] Model performance

[0113] Prediction performance

[0114] If the resulting variable is a binary variable encoded with 0 / 1, the output will be an ROC, a Precision Recall plot, and its data file; if it is a continuous variable, the output will be an EGA (Exploratory Graph Analysis) plot and its data file.

[0115] For continuous independent variables, if you enter "S" in the *Plot column, it means that you have selected the variable. The curve of the relationship between the variable (S) and the predicted value of the outcome variable will be plotted. All other variables will be used as grouping variables (if they are categorical variables, they will be grouped according to their original categories; if they are continuous variables, they will be divided into three groups of low, medium and high based on the 10th, 50th and 90th percentiles) respectively. The curves of S and Y will be plotted separately to observe their interaction with S.

[0116] 2. Research Results

[0117] 2.1 Baseline Characteristics

[0118] At the final survey in August 2022, a total of 595 AIS candidates were recruited for the study. Among these AIS candidates, patients meeting any exclusion criteria were excluded (n=37), and patients with missing data related to NSE, neutrophils, fibrinogen, sex, and age were also excluded from eligible candidates (n=95). Patients with unreliable fibrinogen levels (<0.5 g / L) (n=17) and unreliable neutrophil counts (<1.0 × 10⁹ / L) (n=15) were also excluded from eligible candidates. Therefore, the final analysis included 431 AIS subjects. Based on the clinically normal range of fibrinogen, the 431 AIS subjects were divided into a normal group and a high fibrinogen group. Simultaneously, 89 healthy controls were included in this study. Neutrophil and fibrinogen levels were compared between the 89 healthy controls and the 89 AIS patients using propensity score matching based on age and sex. The study flowchart is shown below. Figure 1 As shown.

[0119] Of the 431 participants in the AIS study, 46.17% (n=199) were female and 53.83% (n=232) were male. The age range of the enrolled participants was 40 to 99 years (female, 49 to 99 years; male, 40 to 91 years), with a mean age of 73.05 ± 10.84 years (female 76.39 ± 10.22 years; male 70.20 ± 10.56 years). The duration of illness before admission ranged from 0.5 to 46 hours, and fibrinogen levels ranged from 1.23 to 6.23 g / L. Baseline characteristics of the enrolled patients are shown in Table 1.

[0120] Table 1. Baseline characteristics of patients grouped according to fibrinogen expression levels

[0121]

[0122]

[0123]

[0124] Comments for English code:

[0125] AFP: Alpha-fetoprotein, ALT: Alanine aminotransferase, BUN: Blood urea nitrogen, CEA: Carcinoembryonic antigen, CA: Cancer, ESR: Erythrocyte sedimentation rate, HCY: Homocysteine, HDL-C: High-density lipoprotein cholesterol, INR: International Normalized Ratio, LACI: Lacunar infarction, LDL-C: Low-density lipoprotein, NSE: Neuron-specific enolase, OCSP: Oxfordshire Community Stroke Project, PACI: Partial anterior circulation infarction, POCI: Posterior circulation infarction, PT: Prothrombin time, PTT: Partial thromboplastin time, RBC: Red blood cells, TACI: Complete anterior circulation infarction; TCH: Total cholesterol, TG: Triglycerides

[0126] Of the 89 healthy controls, 59.55% (n=53) were female and 40.45% (n=36) were male. The age range of the included healthy controls was 48 to 92 years (female, 48–92 years; male, 48–85 years), with a mean age of 68.31 ± 9.01 years (female 69.25 ± 9.69 years; male 66.94 ± 7.83 years). The characteristics of the AIS and healthy controls after propensity score matching by sex and age are shown in Table 2. This indicates that, after propensity score matching, the baseline characteristics of sex and age of the patients were balanced between the two groups.

[0127] Table 2. Characteristics of AIS and healthy control groups after propensity score matching by gender and age.

[0128]

[0129] 2.2 Differential expression levels of NSE, neutrophils, and fibrinogen between the healthy control group and the matched AIS group

[0130] After propensity score matching analysis, there were no differences in sex and age between the healthy control group and the matched AIS group. The mean NSE in the matched AIS group was higher than that in the normal control group (14.07±3.13 g / L vs 12.79±3.70 g / L, p=0.014). Figure 2 A) The fibrinogen level in the AIS-matched group was also higher than that in the healthy control group (2.94±0.81 g / L vs 2.66±0.65 g / L, p=0.012). Figure 2 B).

[0131] 2.3 ROC curve analysis of NSE and fibrinogen between the healthy control group and the matched AIS group

[0132] The AUC (95% confidence interval) of the ROC curves for NSE, neutrophils, and fibrinogen between the healthy control group and the matched AIS group were 0.654 (0.574–0.736, p<0.001), 0.620 (0.537–0.702, p<0.001), and 0.626 (0.543–0.709, p=0.003), respectively. Figure 3 ROC curve analysis showed that the AIS-matched group had higher levels of NSE and fibrinogen compared to the healthy control group.

[0133] 2.4 Correlation analysis of fibrinogen and NSE in healthy control group and matched AIS group

[0134] The Pearson correlation coefficient (95%) between fibrinogen and NSE in the healthy control group was -0.0649 (-0.269 -0.145, p = 0.199), while it was 0.137 (-0.072 -0.336, p = 0.018) in the matched AIS group. This indicates that in the matched AIS group, NSE levels increased with increasing fibrinogen levels, but in the healthy control group, fibrinogen expression levels were not correlated with NSE levels. Figure 4 A, B).

[0135] 2.5 Correlation analysis of fibrinogen and NSE expression levels in AIS patients

[0136] The Person correlation coefficient (95%) between fibrinogen and the central nervous system cell injury marker NSE in the 431 AIS patients included in the study was 0.112 (0.018–0.205, p = 0.019), indicating that in AIS patients, the level of NSE in the central nervous system increases with increasing fibrinogen levels. Figure 5 ).

[0137] 2.6 Univariate analysis of fibrinogen levels in AIS patients

[0138] 431 AIS patients were grouped according to their clinical fibrinogen threshold into a normal fibrinogen group and a high fibrinogen group. Baseline data indicators were compared between the groups (Table 1). Indicators with significant differences between groups (p<0.05), including neutrophils, monocytes, platelets, serum albumin, DD dimer, and coagulation time, as well as routine indicators such as gender and age, were included in univariate analysis. The β value of DD dimer was 0.05 and p=0.05, while the p values ​​of the other indicators were all <0.05. All of these were included as confounding factors in the subsequent curve fitting statistical analysis. The specific results are shown in Table 3 below.

[0139] Table 3 Univariate analysis of factors influencing fibrinogen levels

[0140]

[0141] 2.7 Curve fitting analysis of the relationship between fibrinogen and central nervous system cell damage markers in AIS patients

[0142] Smooth curve fitting analysis: Confounding factors screened by univariate analysis, including monocyte count, platelet count, serum albumin, and clotting time (all p < 0.05), as well as conventional factors such as gender and age, were included as confounding factors in the smooth curve fitting analysis of fibrinogen and NSE. Results showed that after adjusting for gender, age, monocyte count, platelet count, serum albumin, and clotting time, fibrinogen levels showed a linear relationship with NSE with an upward curve. Figure 6 A) Curve fitting analysis using High NSE as the observed outcome revealed a linear relationship of increasing curve between fibrinogen levels and High NSE. Figure 6 B). Smooth curve fitting analysis adjusted for confounding factors showed that the degree of central nervous system cell damage in AIS patients increased with increasing fibrinogen levels.

[0143] 2.8 Establishment and evaluation of a model for predicting post-stroke inflammatory damage using fibrinogen levels

[0144] Logistic regression prediction model establishment: Confounding factors screened by univariate analysis, including monocyte count, platelet count, serum albumin, and clotting time (all p < 0.05), as well as conventional factors such as gender and age, were included in the model analysis for predicting post-stroke inflammatory damage (NSE level) using fibrinogen. Results showed that after adjusting for gender, age, monocyte count, platelet count, serum albumin, and clotting time, the logistic regression model for fibrinogen level predicting post-stroke inflammatory damage was: y = -7.00338 + 5.62873 × (1 / fibrinogen) 2 )-27.20105×[1 / fibrinogen 2 [×log(fibrinogen)] + 0.76015 × (male = 1) + 0.04274 × age + 0.07542 × monocyte count + 0.00343 × platelet count + 0.04120 × albumin + 0.11142 × thrombin time.

[0145] Evaluation of the Logistic Regression Prediction Model: The area under the predicted AUC curve of this model is 0.698 ( Figure 7 The prediction accuracy was 82.8%, with a 95% CI confidence interval of 78.9%–86.3%; the sensitivity was 0.875, and the specificity was 0.972. These results indicate that the Logistic regression model for predicting post-stroke inflammatory damage based on fibrinogen levels has good predictive power.

[0146] Machine learning prediction model establishment and overall evaluation: Confounding factors screened by univariate analysis, including monocyte count, platelet count, serum albumin, and coagulation time (all p < 0.05), as well as conventional factors such as gender and age, were included in the model analysis for predicting post-stroke inflammatory damage using fibrinogen. Results showed that after adjusting for gender, age, monocyte count, platelet count, serum albumin, and coagulation time, the fully automated machine learning prediction model for fibrinogen levels in predicting post-stroke inflammatory damage (…) Figure 8 The results show that: 1) The area under the AUC curve for the test dataset is 0.728, the sensitivity is 0.672, and the accuracy is 0.741; 2) The area under the AUC curve for the train dataset is 0.916, the sensitivity is 0.591, and the accuracy is 0.818; and the area under the AUC curve for the valid dataset is 0.779, the sensitivity is 0.872, and the accuracy is 0.798. These results indicate that the fully automated machine learning prediction model for predicting post-stroke inflammatory damage based on fibrinogen levels has good predictive performance.

[0147] Machine Learning Predictive Model Evaluation – Comparison of Different Algorithm Models: Applicable algorithms were selected based on the type of outcome variables, including deep learning, DRF, GBM, and GLM. Analysis results indicate that the areas under the AUC curves for the four models (deep learning, DRF, GBM, and GLM) predicting post-stroke inflammatory damage based on fibrinogen levels were 0.731, 0.792, 0.728, and 0.685, respectively. Figure 9 The above results show that all four machine learning algorithms achieved good prediction performance, with the DRF model showing the best prediction results.

[0148] 3. Conclusion

[0149] Previous basic experiments have confirmed that fibrin(ogen) deposition in the brain tissue of a mouse MCAO model exacerbates the inflammatory response after stroke. However, in clinical practice, it is difficult to directly monitor the central nervous system-induced inflammatory response and secondary inflammatory damage caused by fibrinogen. Considering the disruption of the blood-brain barrier after AIS, leading to central-peripheral substance exchange, we aim to preliminarily analyze the correlation between changes in fibrinogen in the peripheral circulation of AIS patients and central nervous system injury, providing a potentially convenient monitoring method. In this study, we collected data from 413 AIS patients who met the analytical criteria and analyzed various indicators in their peripheral blood using multiple statistical methods. We first found that, compared to the healthy control group, the levels of fibrinogen and NSE in the peripheral circulation of AIS patients were significantly elevated, and changes in NSE levels were independently correlated with changes in fibrinogen levels. This correlation persisted in analyses after stratified analysis and adjustment for confounding factors, suggesting that during the acute phase of AIS patients, changes in fibrinogen levels in the peripheral circulation are independently correlated with central nervous system damage.

[0150] Previous research has primarily focused on the role of fibrinogen in thrombosis in cardiovascular and peripheral vascular diseases. In cerebrovascular diseases, some studies have preliminarily confirmed the association between fibrinogen and clinical events such as recurrence and prognosis. High fibrinogen levels at admission nearly double the risk of thrombosis after intravenous thrombolysis in AIS patients. Analysis has also shown a correlation between NLR and high fibrinogen in AIS patients undergoing mechanical thrombectomy, which can serve as a predictor of treatment outcome. However, few studies have focused on the role of fibrinogen in the peripheral circulation in the inflammatory response after AIS. Our current study demonstrates that elevated fibrinogen levels are associated with brain inflammatory damage after AIS.

[0151] The mechanism by which high levels of peripheral fibrinogen in the acute phase of acute acute myocardial infarction (AIS) participate in the inflammatory response remains unclear. The vascular, immune, and nervous systems may collectively mediate and influence this interaction. In AIS, neutrophils secrete large amounts of MMP-9, resulting in a significant increase in plasma MMP-9 levels. This neutrophil proliferation also increases the risk of thrombosis. Recent studies have found that platelets interact with P-selectin glycoprotein ligand 1 (PSGL-1) on the neutrophil surface via P-selectin on the platelet surface, triggering NET formation, activating the intrinsic coagulation pathway, promoting the coagulation cascade, and degrading inhibitors of the natural anticoagulant tissue factor (TF) pathway, further exacerbating thrombosis. On the other hand, elevated peripheral fibrinogen levels have also been shown to be associated with the development of inflammatory diseases, playing a significant role in severe and chronic low-grade inflammation. This suggests that the interaction between peripheral fibrinogen and various immune cells plays a driving role in disease development. Our previous studies have also confirmed that fibrin(ogen) deposited in MCAO brain tissue mainly works by binding to integrin receptor α on the surface of neutrophils. M β2 activates neutrophils.

[0152] Inflammatory factors in ischemic lesions originate from endogenous damaged nerve tissue and exogenous blood circulation. The latter includes infiltrating leukocytes and fibrinogen, both of which have been observed in ischemic lesions in previous studies. Infiltrating leukocytes are recruited to the infarcted area and, after activation, amplify the inflammatory cascade response following acute ischemic infarction by secreting a variety of pro-inflammatory mediators. In short, after ischemic stroke, inflammatory effector factors in the peripheral circulation leak into the ischemic lesion through multiple pathways and act as inflammatory mediators to promote neuronal apoptosis and aggravate secondary brain injury. In recent years, NSE has been used as a stable and reliable biochemical indicator of neuronal damage in AIS patients. NSE is a glycolytic enzyme that participates in the anaerobic glycolysis process after ischemic stroke. Previous studies have shown that the cell integrity of central neurons is disrupted under hypoxic conditions, and NSE is released into the intercellular space and cerebrospinal fluid. Disorders of energy metabolism in glial cells after acute ischemic stroke also lead to elevated serum NSE expression

[25] . Because the blood-brain barrier is disrupted during acute ischemia, non-encephalocyte serous ethers (NSEs) enter the circulatory system, thereby increasing the expression level of NSEs in the circulation of AIS patients. Secondary brain injury caused by the inflammatory response following stroke further promotes neuronal apoptosis, accelerating the release of NSEs from the central nervous system into the peripheral blood circulation. The results suggest that, compared to healthy controls, AIS patients had significantly elevated levels of both fibrinogen and NSE, with a linear correlation between the two.

[0153] Subjects with underlying diseases that could affect fibrinogen levels, such as malignancies and hematologic disorders, were excluded. Multiple analytical methods were employed to ensure the reliability of the results. In this study, we did not use the NIHSS score to assess the severity of central neuronal injury. This is because some ischemic lesions can occur in asymptomatic non-functional areas, thus introducing a low NIHSS score bias. Conversely, the NSE directly reflects the damage to neurons in both functional and non-functional areas. Therefore, using the NSE to assess the severity of central neuronal injury ensures the scientific validity of the conclusions.

[0154] In conclusion, the model based on fully automated machine learning to predict post-stroke inflammatory damage at fibrinogen levels has good predictive ability and can be used clinically to predict brain tissue inflammatory damage after cerebral infarction.

[0155] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention, enabling those skilled in the art to understand and apply it. However, it should not be construed that the specific implementation of the invention is limited to these descriptions. For those skilled in the art, several simple deductions or substitutions can be made without departing from the inventive concept, without requiring creative effort. Therefore, any simple improvements made to the present invention by those skilled in the art based on the disclosure thereof should be within the scope of protection of the present invention.

Claims

1. A method for establishing a predictive model for post-stroke inflammatory damage, characterized in that, The establishment method includes the following steps: Step 1: Collect clinical indicator data from the AIS group and the healthy control group; Step 2: After matching the AIS group and the healthy control group according to their propensity scores based on gender and age, a matched AIS group with the same number of members as the healthy control group is obtained, and the matching characteristics between the healthy control group and the matched AIS group are obtained. Step 3: Analysis of differences and correlations in matching characteristics between the healthy control group and the matched AIS group; Step 4: Based on the analysis in Step 3, conduct a univariate analysis of the influencing factors of fibrinogen levels in the AIS group; Step 5: Input the confounding factors identified in Step 4 into the fully automated machine learning model for model training and validation to obtain a Logistic regression model for predicting post-stroke inflammatory damage based on fibrinogen levels.

2. The method for establishing a post-stroke inflammatory injury prediction model according to claim 1, characterized in that: The clinical data collected in step 1 include gender, age, past medical history, complete blood count, serum biochemical indicators, blood lipid indicators, fibrinogen level, D-dimer, partial thromboplastin time, prothrombin time, NSE, tumor markers, and treatment received before enrollment.

3. The method for establishing a post-stroke inflammatory injury prediction model according to claim 2, characterized in that, Past medical history includes hypertension, coronary heart disease, diabetes, and atrial fibrillation; routine blood tests include red blood cell count, white blood cell count, lymphocyte count, neutrophil count, and platelet count.

4. The method for establishing a post-stroke inflammatory injury prediction model according to claim 3, characterized in that, The matching features obtained in step 2 include gender, age, white blood cell count, neutrophil count, fibrinogen level, NSE, hypertension, diabetes, coronary heart disease, and atrial fibrillation.

5. The method for establishing a post-stroke inflammatory injury prediction model according to claim 4, characterized in that, Step 3 includes the following steps: Step 301: Analysis of differential expression levels of NSE, neutrophils, and fibrinogen between the healthy control group and the matched AIS group; Step 302: ROC curve analysis of NSE and fibrinogen between the healthy control group and the matched AIS group; Step 303: Correlation analysis of fibrinogen and NSE between the healthy control group and the matched AIS group.

6. The method for establishing a post-stroke inflammatory injury prediction model according to claim 5, characterized in that, Before conducting univariate analysis of fibrinogen levels in AIS patients, it is necessary to analyze the correlation between fibrinogen and NSE expression levels in AIS patients and determine the dependence of fibrinogen and NSE expression levels in AIS patients.

7. The method for establishing a post-stroke inflammatory injury prediction model according to claim 6, characterized in that, Confounding factors identified by the univariate analysis in step 4 included sex, age, neutrophils, monocytes, platelets, serum albumin, DD dimer, and thrombin time.

8. The method for establishing a post-stroke inflammatory injury prediction model according to claim 7, characterized in that, After the univariate analysis in step 4, a curve fitting analysis was performed to determine the relationship between fibrinogen and central nervous system cell damage markers in AIS patients, as detailed below: Confounding factors screened by univariate analysis were incorporated into the smooth curve fitting analysis of fibrinogen and NSE to determine that the degree of central nervous cell damage in AIS patients increased with increasing fibrinogen levels.

9. The method for establishing a post-stroke inflammatory injury prediction model according to claim 8, characterized in that, The logistic regression model established in step 5 for predicting post-stroke inflammatory damage based on fibrinogen levels is as follows: y = -7.00338 + 5.62873 × (1 / fibrinogen) 2 )-27.20105×[1 / fibrinogen 2 [×log(fibrinogen)] + 0.76015 × (male = 1) + 0.04274 × age + 0.07542 × monocyte count + 0.00343 × platelet count + 0.04120 × albumin + 0.11142 × thrombin time.

10. A method for assessing the degree of post-stroke inflammatory damage using a post-stroke inflammatory damage prediction model established according to any one of claims 1-9, characterized in that, Includes the following steps: (1) Collect the patient's fibrinogen level, monocyte count, platelet count, albumin count and thrombin time; (2) The collected fibrinogen level, monocyte count, platelet count, albumin count, thrombin time, and the patient's age and gender were input into the Logistic regression model to calculate the predicted value of post-stroke inflammatory damage.

Images