A method and system for predicting response to treatment in patients with chronic hepatitis b

Abstract

This invention discloses a method and system for predicting treatment response in patients with chronic hepatitis B, addressing the technical problem that existing methods for predicting treatment response in chronic hepatitis B patients cannot meet the needs of early clinical warning. The method includes acquiring and preprocessing raw medical data to obtain a regularized clinical dataset; calculating the individualized HBsAg decline slope and the total HBsAg decline slope during the target treatment phase using HBsAg dynamic slope calculation; combining the patient's actual response labels to complete model construction and optimization, obtaining the optimal treatment response prediction model and defining the best threshold for response prediction and a high-sensitivity warning threshold. After receiving the medical data to be tested, the method generates patient treatment response probability determination results and risk stratification early warning results, respectively; finally, it integrates these two types of results to output the patient treatment response prediction result.

Description

Technical Field

[0001] This invention relates to the field of medical data analysis technology, and in particular to a method and system for predicting treatment response in patients with chronic hepatitis B. Background Technology

[0002] Currently, the treatment goals for chronic hepatitis B (CHB) have gradually expanded from simply suppressing viral replication to pursuing deeper virological and serological responses, such as significant reductions in HBsAg, HBsAg clearance, and functional cure. Nucleos(t)ide analogues (NAs) can effectively suppress HBV (chronic hepatitis B virus) replication in the long term, delaying the progression of liver inflammation and fibrosis, and reducing the risk of HCC. In addition, pegylated interferon (PEG-IFN) and some novel deantigenation treatment strategies also aim to promote HBsAg reduction and increase the probability of functional cure. The dynamic changes in HBsAg, especially the rate and magnitude of its decline, are considered important biomarkers reflecting the disease process, treatment response, and long-term outcome.

[0003] In CHB treatment studies, week 48 is often used as the standard time point for efficacy assessment, but there is no universally accepted definition of "48-week response." Different studies use virological suppression, serological conversion, HBsAg clearance, or a significant decrease in HBsAg as outcome indicators. In recent years, a decrease in HBsAg ≥1 log10 IU / mL has gradually become a commonly used interim response endpoint in studies related to functional cure, especially in clinical trials of PEG-IFNadd-on, novel siRNAs, immunomodulators, and combination therapies. Compared with final HBsAg clearance, this endpoint has a higher event rate, earlier occurrence, and higher statistical efficiency, making it suitable as a surrogate or intermediate endpoint in efficacy evaluation, patient screening, and dynamic prediction studies.

[0004] Existing methods for predicting treatment response in patients with chronic hepatitis B largely rely on single baseline indicators, endpoint response results, or complex longitudinal trajectory modeling throughout the treatment process. These methods construct predictive models using traditional statistical or machine learning algorithms, focusing on the overall prediction of the final treatment outcome. However, these methods fail to identify key early warning windows during treatment or extract interpretable core parameters of early dynamic changes. They struggle to capture key features reflecting subsequent response potential in the early stages of treatment and cannot quickly and easily stratify patients for early risk. Therefore, existing methods for predicting treatment response in patients with chronic hepatitis B cannot meet the clinical need for early warning. Summary of the Invention

[0005] This invention provides a method and system for predicting treatment response in patients with chronic hepatitis B, which solves the technical problem that existing methods for predicting treatment response in patients with chronic hepatitis B cannot meet the needs of early clinical warning.

[0006] The first aspect of this invention provides a method for predicting treatment response in patients with chronic hepatitis B, comprising: Obtain raw medical records and preprocess them to obtain a regularized clinical dataset. The dynamic slope of HBsAg was calculated on the regularized clinical dataset to obtain the individualized HBsAg decline slope for each patient in multiple treatment stages and the total HBsAg decline slope for each patient in the target treatment stage. The optimal treatment response prediction model is obtained by optimizing the model based on the patient's actual response label, the individualized HBsAg decline slope of each patient in multiple treatment stages, and the regularized clinical dataset. Based on the slope of the total HBsAg decrease for each patient during the target treatment phase and the patient’s actual response label, a prediction threshold is defined to obtain the optimal response prediction threshold and the high-sensitivity warning threshold. When the medical data to be tested is received, the optimal treatment response prediction model is used to predict the treatment response probability of each patient in the future treatment stage and judge it one by one with the preset probability threshold, and output the patient treatment response probability judgment result. Based on the slope of the decrease in total HBsAg in the target treatment stage for each patient corresponding to the medical data to be tested, the optimal threshold for response prediction, and the high-sensitivity early warning threshold, the patient treatment response risk stratification and early warning determination are performed to obtain the patient personalized risk stratification early warning results. By integrating the patient's treatment response probability determination results and the patient's personalized risk stratification early warning results, the patient's treatment response prediction results are output.

[0007] Optionally, the preprocessing of the original medical records to obtain a regularized clinical dataset includes: The original medical records are filtered to output a cohort of patients receiving treatment for chronic hepatitis B. Treatment data were collected from the cohort of patients with chronic hepatitis B and the raw clinical dataset was output. The original clinical dataset is processed sequentially with variable unification, outlier correction, logarithmic transformation, and missing value imputation to output a normalized clinical dataset.

[0008] Optionally, the step of calculating the dynamic slope of HBsAg on the regularized clinical dataset to obtain the individualized HBsAg decline slope for each patient in multiple treatment phases and the total HBsAg decline slope for each patient in the target treatment phase includes: The individualized segmented HBsAg decline slope is fitted and calculated on the regularized clinical dataset, and the individualized HBsAg decline slope for each patient in multiple treatment stages is output. The individualized HBsAg reduction slopes of each patient at each treatment stage were integrated to obtain the total HBsAg reduction slope for each patient at the target treatment stage.

[0009] Optionally, the step of optimizing the model based on the patient's actual response label, the individualized HBsAg decline slope for each patient across multiple treatment stages, and the regularized clinical dataset to obtain the optimal treatment response prediction model includes: Feature extraction is performed on the regularized clinical dataset to output the baseline clinical characteristics and treatment plan characteristics of each patient. The baseline clinical characteristics, treatment plan characteristics, and individualized HBsAg decline slopes at multiple treatment stages of each patient are matched, merged, and standardized according to the patient's unique ID to form a model input feature matrix with one record per line for each patient. The actual response labels of the patients are used as training labels to construct a modeling feature dataset. The modeling feature dataset is randomly divided to output a model training set and a model test set. The preset machine learning model is subjected to hyperparameter optimization and model training on the model training set, and multiple candidate prediction models are output. Multiple candidate prediction models are validated on the model test set, and the model with the best performance is selected and the optimal treatment response prediction model is output.

[0010] Optionally, the step of defining a prediction threshold based on the slope of the total HBsAg decrease for each patient during the target treatment phase and the patient's actual response label to obtain an optimal response prediction threshold and a high-sensitivity warning threshold includes: Based on the total HBsAg decrease slope of each patient in the target treatment phase and the actual response label of the patient, the performance index of the candidate cutoff point is calculated, and multiple candidate cutoff point performance indices are output. Optimal interception points are selected from multiple candidate interception point performance metrics, and the optimal threshold for response prediction is output. The candidate cutoff point performance indicators are initially screened according to a preset sensitivity, and then the cutoff point with the highest specificity is selected from the initial screening results to output a high-sensitivity early warning threshold.

[0011] Optionally, the step of using the optimal treatment response prediction model to predict based on the medical data to be tested, outputting the treatment response probability of each patient in the future treatment stage, and judging it one by one with a preset probability threshold, and outputting the patient treatment response probability judgment result, includes: The medical data to be tested is preprocessed to obtain a regularized clinical dataset. The individualized segmented HBsAg decline slope of the measured regular clinical dataset is fitted and calculated, and the individualized HBsAg decline slope of each patient in multiple treatment stages is output. Feature extraction is performed on the regular clinical dataset to be tested, and the baseline clinical features and treatment plan features of each patient are output. The baseline clinical characteristics to be tested, the characteristics of the treatment plan to be tested, and the individualized HBsAg decline slope to be tested in multiple treatment stages of each patient are respectively input into the optimal treatment response prediction model for prediction, and the probability of treatment response for each patient in future treatment stages is output. The probability of each patient's treatment response in the future treatment stage is determined against a preset probability threshold, and the result of the determination of the patient's treatment response probability is output.

[0012] Optionally, the step of performing patient treatment response risk stratification and early warning determination based on the decline slope of total HBsAg in the target treatment phase for each patient corresponding to the test medical data, the optimal response prediction threshold, and the high-sensitivity early warning threshold, to obtain personalized risk stratification early warning results for patients, includes: The slope of the decrease in total HBsAg during the target treatment phase for each patient is compared with the optimal threshold for response prediction. Patients whose slope of the decrease in total HBsAg reaches or falls below the optimal threshold for response prediction are classified into the high-probability response group, while patients whose slope of the decrease in total HBsAg does not reach the optimal threshold for response prediction are classified into the high-risk non-response group. The initial grouping results are then output. The slope of the decrease in total HBsAg during the target treatment phase of each patient is compared with the high-sensitivity warning threshold. Patients whose slope of the decrease in total HBsAg reaches the high-sensitivity warning threshold are marked, and the early warning marking results are output. The initial grouping results and the early warning labeling results are integrated to output personalized risk stratification early warning results for patients.

[0013] A second aspect of the present invention provides a treatment prediction response system for patients with chronic hepatitis B, comprising: The acquisition module is used to acquire raw medical data and preprocess the raw medical data to obtain a regularized clinical dataset. The calculation module is used to calculate the dynamic slope of HBsAg on the regularized clinical dataset to obtain the individualized HBsAg decline slope for each patient in multiple treatment stages and the total HBsAg decline slope for each patient in the target treatment stage. The construction module is used to optimize the model based on the patient's actual response label, the individualized HBsAg decline slope of each patient in multiple treatment stages, and the regularized clinical dataset, so as to obtain the optimal treatment response prediction model. The definition module is used to define the prediction threshold based on the slope of the total HBsAg decrease for each patient in the target treatment phase and the patient’s actual response label, so as to obtain the optimal response prediction threshold and the high-sensitivity warning threshold. The prediction module is used to predict the treatment response probability of each patient in the future treatment stage by using the optimal treatment response prediction model when receiving the medical data to be tested, and to judge the patient's treatment response probability one by one with the preset probability threshold. The prediction module is used to output the patient's treatment response probability judgment result. The determination module is used to perform patient treatment response risk stratification and early warning determination based on the decline slope of the total HBsAg to be tested for each patient in the target treatment stage, the optimal threshold for response prediction, and the high-sensitivity early warning threshold, so as to obtain the patient's personalized risk stratification early warning result. The output module is used to integrate the patient treatment response probability determination results and the patient personalized risk stratification early warning results to output the patient treatment response prediction results.

[0014] A third aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the treatment prediction response method for patients with chronic hepatitis B as described above.

[0015] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the treatment prediction response method for patients with chronic hepatitis B as described above.

[0016] As can be seen from the above technical solutions, the present invention has the following advantages: The above-mentioned technical solution of the present invention provides a method for predicting treatment response in patients with chronic hepatitis B. The method involves acquiring raw medical data and preprocessing it to obtain a regularized clinical dataset. HBsAg dynamic slope calculation is performed on the regularized clinical dataset to obtain the individualized HBsAg decline slope for each patient across multiple treatment stages and the total HBsAg decline slope for each patient in the target treatment stage. Based on the patient's actual response label, the individualized HBsAg decline slope for each patient across multiple treatment stages, and the regularized clinical dataset, a model is constructed and optimized to obtain an optimal treatment response prediction model. A prediction threshold is defined based on the total HBsAg decline slope for each patient in the target treatment stage and the patient's actual response label to obtain an optimal response prediction threshold and a high-sensitivity warning threshold. When medical data to be tested is received, the optimal treatment response prediction model is used to predict the response based on the medical data, outputting the treatment response probability for each patient in the future treatment stage and comparing it with a preset probability threshold to determine the patient's treatment response probability. The final result is then determined based on the corresponding medical data. For each patient, the treatment response risk is stratified and early warning is determined by the slope of the decrease in total HBsAg, the optimal threshold for response prediction, and the high-sensitivity warning threshold during the target treatment phase. This yields personalized risk stratification and early warning results. The treatment response probability determination results and personalized risk stratification and early warning results are integrated to output the patient's treatment response prediction results. Based on the above, this invention relies on the dynamic HBsAg decrease characteristics in the early stages of treatment to conduct quantitative analysis and model training. This fully explores the efficacy prediction value of early treatment indicators, overcoming the shortcomings of traditional prediction methods, such as strong lag and inability to make early judgments. By establishing a dual judgment system that combines machine learning multi-feature comprehensive prediction with slope threshold stratification and early warning, the system not only uses model fusion of multi-dimensional early data to ensure the overall accuracy of response prediction, but also achieves rapid risk stratification and early warning based on the early antigen decrease trend through exclusive quantitative thresholds. This can accurately identify the long-term response risk of patients in the early stages of treatment, enabling early prediction and risk warning of treatment response in patients with chronic hepatitis B, and meeting the needs of early clinical diagnosis and treatment assessment. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the steps in a treatment response prediction method for patients with chronic hepatitis B provided in Embodiment 1 of the present invention; Figure 2This is a flowchart of the technology for predicting treatment response at week 48 based on the dynamic decline slope of HBsAg 24 weeks before treatment, as provided in Embodiment 1 of the present invention. Figure 3 This is a structural block diagram of a treatment prediction and response system for patients with chronic hepatitis B, provided in Embodiment 2 of the present invention. Detailed Implementation

[0019] This invention provides a method and system for predicting treatment response in patients with chronic hepatitis B, which solves the technical problem that existing methods for predicting treatment response in patients with chronic hepatitis B cannot meet the needs of early clinical warning.

[0020] 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, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be noted that in the optional embodiments of the present invention, the object information and other related data involved require the permission or consent of the object when the embodiments of the present invention are applied to specific products or technologies, and the collection, use, and processing of related data must comply with relevant laws, regulations, and standards. That is to say, if the embodiments of the present invention involve data related to the object, it needs to be obtained with the authorization and consent of the object, the authorization and consent of relevant departments, and in compliance with relevant laws, regulations, and standards. If personal information is involved in the embodiments, the acquisition of all personal information requires the consent of the individual. If sensitive information is involved, the separate consent of the information subject is required, and the embodiments also need to be implemented with the authorization and consent of the object.

[0021] Terminology Explanation: HBV DNA: Hepatitis B virus deoxyribonucleic acid load, used to reflect the level of viral replication.

[0022] BMI: Body Mass Index.

[0023] A decrease in HBsAg ≥ 1 log10 IU / mL means that the quantitative value of HBsAg has decreased by 1 logarithmic unit to the base 10 compared with the baseline, which is equivalent to a decrease of about 10 times in HBsAg.

[0024] Functional cure: usually refers to the continuous seroconversion of HBsAg, with or without the appearance of anti-HBs, but it is not the same as the complete elimination of the virus.

[0025] log10(HBsAg): The result of performing a base-10 logarithmic transformation on the HBsAg value, used to reduce the impact of skewed distribution on statistical modeling.

[0026] Longitudinal data: refers to data collected repeatedly from the same patient at multiple time points.

[0027] Dynamic trajectory: refers to the continuous process or trend of a certain indicator changing over treatment time.

[0028] Slope: refers to the average rate of change of a certain indicator per unit time. In this invention, it is selected as the estimated slope of log10(HBsAg) changing with treatment time.

[0029] Individualized decline slope: refers to the patient-specific decline rate estimated based on the overall model and combined with the patient's own follow-up data.

[0030] Piecewise linear mixed-effects model: A statistical model that simultaneously considers the overall average trend and individual differences, used to extract individualized decline slopes for different time periods.

[0031] Fixed effects: Parameters in the model that represent the general trend of change in the population.

[0032] Random effects: Parameters in a model that represent the degree to which an individual deviates from the population average.

[0033] Kaplan-Meier curve: A statistical method used to describe the distribution of events over time.

[0034] The log-rank test is a statistical test used to compare differences in the time distribution of events between different groups.

[0035] Linear regression: A statistical method used to analyze the relationship between continuous outcomes and influencing factors.

[0036] Logistic regression: A statistical method used to analyze the relationship between binary outcomes and influencing factors.

[0037] Markov chain transition: refers to the process by which a patient changes from one response state to another between different stages of treatment.

[0038] Machine learning model: refers to an algorithmic model that is trained using data and is capable of classification or prediction.

[0039] XGBoost: Extreme Gradient Boosting, is a commonly used machine learning classification and regression model.

[0040] Random Forest: A machine learning approach based on the ensemble of multiple decision trees.

[0041] Training set: A dataset used for model training and parameter learning.

[0042] Test set: An independent dataset used to evaluate model performance.

[0043] Ten-fold cross-validation: A model evaluation method that divides the training data into 10 parts and performs training and validation in turn.

[0044] Grid search: A method to find the optimal model parameters by traversing preset combinations of parameters.

[0045] ROC curve: Receiver Operating Characteristic Curve, used to evaluate the model's discriminative ability.

[0046] AUC: Area Under the Curve, used to quantify the overall discriminative power of a model.

[0047] Calibration curve: A graphical method used to evaluate the consistency between the model's predicted probability and the actual outcome.

[0048] DCA: Decision Curve Analysis, used to evaluate the clinical net benefit of a model.

[0049] Bootstrap resampling: a method for evaluating model stability by repeatedly sampling with replacement.

[0050] Sensitivity: refers to the proportion of real respondents who are correctly identified.

[0051] Specificity refers to the proportion of real nonresponders that are correctly excluded.

[0052] Youden index: defined as Sensitivity + Specificity - 1, used to comprehensively evaluate classification performance at different thresholds.

[0053] Risk stratification: The process of classifying patients into high-probability response groups, high-risk groups, or other risk levels based on prediction results or thresholds.

[0054] High-probability response group: refers to the patient group that is judged by the prediction rules of this invention to have a high probability of subsequently reaching the preset therapeutic endpoint.

[0055] High-risk group / low-response-risk group: refers to the patient group that is judged by the prediction rules of this invention to have a low probability of reaching the preset therapeutic endpoint.

[0056] Please see Figure 1 , Figure 1 This is a flowchart illustrating the steps of a treatment response prediction method for patients with chronic hepatitis B, as provided in Embodiment 1 of the present invention.

[0057] This invention provides a method for predicting treatment response in patients with chronic hepatitis B, comprising: Step 101: Obtain the original medical records and preprocess them to obtain a regularized clinical dataset.

[0058] Original medical data refers to the original medical information of patients with chronic hepatitis B collected from the clinical diagnosis and treatment system without any processing, covering all original medical data such as patient basic information, treatment-related information, and laboratory test information.

[0059] It should be noted that the original medical data of patients with chronic hepatitis B are retrieved from the clinical diagnosis and treatment system. Preprocessing operations such as screening, format normalization, and anomaly correction are carried out on the original medical data according to the established processing standards to form a normalized clinical dataset that can be used for subsequent analysis.

[0060] It is worth mentioning that the content of the treatment data collection in this invention includes: (1) General characteristics: age, gender, BMI; (2) Treatment plan; (3) Past medication history; (4) Family history of illness; (5) Disease diagnosis staging; (6) Laboratory test data: including baseline HBsAg, HBeAg, HBV DNA, Anti-HBs, ALT, neutrophil count, and weekly changes in HBsAg titer during treatment; (7) Lifestyle information, such as smoking and drinking habits.

[0061] Further, step 101 may include the following sub-steps: S11. Filter the original medical data and output the chronic hepatitis B treatment patient cohort; S12. Collect treatment data from the cohort of patients with chronic hepatitis B and output the original clinical dataset; S13. Perform variable unification, outlier correction, logarithmic transformation, and missing value imputation on the original clinical dataset in sequence, and output a normalized clinical dataset.

[0062] It should be noted that the process involves obtaining and filtering the original medical records to output a cohort of patients undergoing chronic hepatitis B treatment. In this embodiment, original medical records related to chronic hepatitis B were extracted from the electronic medical record system of the Seventh Affiliated Hospital of Sun Yat-sen University in Shenzhen from May 2018 to December 2025. Patient screening was completed item by item according to the established inclusion and exclusion criteria. The inclusion criteria limited the subjects to being between 18 and 65 years old, having been infected with hepatitis B virus for at least 6 months, and not having received pegylated interferon. The study included subjects with contraindications to medication, a minimum of 48 weeks of standardized treatment, and excluded individuals with concurrent infections of hepatitis C virus, human immunodeficiency virus, or other pathogens, as well as those diagnosed with decompensated cirrhosis, liver cancer, or other end-stage liver diseases. After screening, a cohort of 513 qualified patients with chronic hepatitis B was generated. Treatment data was collected from this cohort, and a raw clinical dataset was output. This embodiment uses the selected patient cohort as the data collection target, comprehensively collecting basic personal information, interferon monotherapy or combination / sequential therapy regimens, past medical history, family history of liver disease, clinical stage of the disease, various laboratory test results, and daily life-related information. All collected information was aggregated and integrated to generate a raw clinical dataset without standardization. The raw clinical dataset underwent variable unification, outlier correction, logarithmic transformation, and missing value imputation to output a standardized clinical dataset. This embodiment first performs a variable unification operation, unifying the variable names, units of measurement, and data storage formats of all fields in the dataset. Then, outlier correction is performed, screening outliers in the data that deviate from the clinically reasonable range and correcting them in accordance with medical standards. Next, a logarithmic transformation with base 10 is performed on the skewed HBsAg index to obtain log10(HBsAg) format data to complete the logarithmic transformation. Finally, appropriate statistical methods such as multiple imputation and nearest neighbor imputation are used to fill blank entries to achieve missing value imputation. After all processing steps are completed, a standardized and well-organized clinical dataset is output.

[0063] Step 102: Calculate the dynamic slope of HBsAg on the regularized clinical dataset to obtain the individualized HBsAg decline slope for each patient in multiple treatment stages and the total HBsAg decline slope for each patient in the target treatment stage.

[0064] The individualized HBsAg decline slope refers to the slope value obtained by trend fitting calculation based on the time-series detection data of hepatitis B surface antigen (HBsAg) continuously collected within two independent treatment segments of 0-12 weeks and 12-24 weeks. This indicator can accurately quantify the average weekly decline rate of hepatitis B surface antigen in a single patient within the corresponding single treatment segment. It is also the core feature data used in this invention to build predictive models and predict the probability of treatment response.

[0065] The target treatment phase is a core observation period specifically defined in this invention for early efficacy assessment, specifically limited to 0-24 weeks after the patient receives treatment. This phase is further divided into two consecutive sub-treatment segments: 0-12 weeks and 12-24 weeks. All key operations in this invention, such as HBsAg dynamic slope calculation, prediction threshold definition, risk stratification, and early warning, are completed based on clinical data collected within this phase.

[0066] The total HBsAg decline slope refers to the comprehensive slope value obtained by integrating the individualized HBsAg decline slopes corresponding to the two treatment segments of 0-12 weeks and 12-24 weeks for a single patient. This indicator is generated based on all HBsAg test data of the patient during the complete 0-24 week target treatment phase. It is used to uniformly quantify the overall average decline rate of hepatitis B surface antigen in a single patient during the entire 0-24 week treatment cycle. It is mainly used in the threshold definition, risk stratification and early warning judgment of this invention.

[0067] It should be noted that the standardized HBsAg time-series detection data of each patient in the regularized clinical dataset were retrieved, and the dynamic change trend of HBsAg in the two treatment segments of 0-12 weeks and 12-24 weeks were fitted respectively. The individualized HBsAg decline slope corresponding to each segment was calculated, and then the slope data of the two segments were integrated to calculate the total HBsAg decline slope of the patient in the target treatment stage of 0-24 weeks.

[0068] Furthermore, step 102 may include the following sub-steps: S21. Perform individualized segmented HBsAg decline slope fitting calculation on the regular clinical dataset, and output the individualized HBsAg decline slope for each patient in multiple treatment stages. S22. Integrate the individualized HBsAg decline slope of each patient at each treatment stage to obtain the total HBsAg decline slope of each patient at the target treatment stage.

[0069] It should be noted that this invention includes a Kaplan-Meier curve module to describe the time distribution for achieving the preset HBsAg response outcome, and uses a log-rank test to compare differences between different groups. Simultaneously, linear regression and logistic regression modules are included to analyze risk factors affecting HBsAg response time, response status, and response speed.

[0070] In addition, a Markov chain analysis module is set up to characterize the dynamic transition process of the patient's HBsAg response status and the rate of decrease in hepatitis B surface antigen between different treatment stages. Preferably, the treatment stages are divided into 0-12 weeks, 12-24 weeks, 24-36 weeks, and 36-48 weeks, and the state transition patterns between each stage are analyzed.

[0071] Furthermore, to quantify the dynamic trajectory of HBsAg changes in patients over the 24 weeks prior to treatment, this invention preferably employs a piecewise linear mixed-effects model to extract the individualized dynamic decline slope. The model is as follows: HbsAg_log10= ; in: 1 represents the overall average slope from 0 to 12 weeks, which is the overall average decrease slope of log10 (HBsAg) in the subject population during the 0-12 week treatment phase; 2 represents the overall mean slope from week 12 to week 24, i.e., the overall mean decline slope of log10(HBsAg) in the study population during the 12-24 week treatment phase; u1i represents the individual deviation of the i-th patient relative to the overall mean, i.e., the individualized decline slope of the i-th patient from week 0 to week 12 relative to the population mean. Individual offset parameter of 1; The fixed intercept term of the model represents the mean HbsAg_log10 level of all patients at the initial stage of treatment (baseline), i.e., the baseline HbsAg_log10 value without considering time effects and individual differences; For the i-th patient, the time measurement independent variable for the 0-12 week treatment phase; For the i-th patient, the time measurement independent variable is the treatment phase from 12 to 24 weeks. is the individualized baseline random offset of the i-th patient relative to the baseline mean of the subject population; The individualized decline slope of patient i from week 12 to 24 relative to the population mean Individual offset parameters of 2; This is the random error term of the model, representing the random data deviation in the sample detection data that cannot be fitted by the various parameters of the model.

[0072] Therefore, the individualized rate of decline for each patient at different stages can be obtained: 0-12 cycle rate = b1 + u1; 12-24 cycle rate = b2 + u2; That is, the estimated slope of the change in log10(HBsAg) with weekly treatment time for each patient within the corresponding time period.

[0073] Where b1 represents the overall mean decrease slope of log10(HBsAg) in all subjects during the 0-12 week treatment phase, and parameters 1. One-to-one correspondence; u1 is the individual offset of the decline slope of the i-th patient during the 0-12 week phase relative to the population average b1; b2 is the overall average decline slope of log10 (HBsAg) for all subjects during the 12-24 week treatment phase, and parameters 2-to-one correspondence; u2 is the individual offset of the decline slope of the i-th patient in the 12-24 week phase relative to the population average level b2; Based on the above, time-series detection data of hepatitis B surface antigen (HBsAg) for each patient were extracted from the regularized clinical dataset. All valid detection values ​​were logarithmically transformed to base 10 to construct a log10 (HBsAg) quantified sequence. A piecewise linear mixed-effects model was then used, dividing the first 24 weeks of treatment into two consecutive treatment segments: 0-12 weeks and 12-24 weeks. Patients were included in the model fitting. In the model, time-evaluation independent variables t1 and t2 were set for each detection time segment. t1 was set to the corresponding week number when the detection time was between 0 and 12 weeks, and 0 when the detection time exceeded 12 weeks. t2 was set to the week number between "detection time - 12 weeks" when the detection time was between 12 and 24 weeks, and 0 when the detection time was less than 12 weeks. Simultaneously, a sample indicator variable cij was used to indicate whether each detection data point was a valid analysis sample. The log10 (HBsAg) sequence for each patient was used as the dependent variable, combined with the fixed-effects slope at the population level. 1. 2. The individual-level random bias terms u1i and u2i are fitted together. After fitting, the individualized HBsAg decrease slope for each patient during the 0-12 week and 12-24 week treatment phases is calculated, i.e., the individualized decrease rate during the 0-12 week phase is... The sum of 1 and u1i, the rate of decrease in individualization from week 12 to week 24 is The sum of 2 and u2i; after completing the segmented slope fitting, the individualized HBsAg decline slopes of each patient in the 0-12 week and 12-24 week phases are integrated and calculated to obtain the total HBsAg decline slope of the patient in the target treatment phase of 0-24 weeks. Finally, the multi-stage individualized HBsAg decline slopes and the total HBsAg decline slopes in the target treatment phases of all patients are output.

[0074] Step 103: Optimize the model based on the patient's actual response labels, the individualized HBsAg decline slope for each patient across multiple treatment stages, and the regularized clinical dataset to obtain the optimal treatment response prediction model.

[0075] It should be noted that the baseline clinical characteristics and treatment plan characteristics of each patient are extracted from the regular clinical dataset, matched and merged with the individualized HBsAg decline slope at each stage according to the patient number and standardized, and the modeling dataset is constructed using the actual response label of the patient as the training label. The dataset is divided into training set and test set, and the preset machine learning model is trained, hyperparameters are optimized and validated. The model with the best performance is selected to obtain the optimal treatment response prediction model.

[0076] Furthermore, step 103 may include the following sub-steps: S31. Extract features from the regularized clinical dataset and output the baseline clinical features and treatment plan features of each patient; S32. The baseline clinical characteristics, treatment plan characteristics, and individualized HBsAg decline slope of each patient at multiple treatment stages are matched, merged, and standardized according to the patient's unique number to form a model input feature matrix with one record per line for each patient. The actual response labels of the patients are used as training labels to construct the modeling feature dataset. S33. Randomly divide the modeling feature dataset and output the model training set and model test set; S35. Perform hyperparameter optimization and model training on the model training set for the preset machine learning model, and output multiple sets of candidate prediction models. S36. Perform model validation on the model test set for multiple candidate prediction models, select the model with the best performance, and output the optimal treatment response prediction model.

[0077] The patient’s actual response label was a decrease of ≥1 log10 IU / mL in HBsAg from baseline at week 48.

[0078] It should be noted that in terms of model construction, variables such as baseline hepatitis B surface antigen, age, hepatitis B virus DNA load, baseline hepatitis B core antigen, neutrophil count, and alanine aminotransferase were first included, and then the rate of decline of hepatitis B surface antigen at 0-12 weeks and 12-24 weeks were further included as dynamic features.

[0079] Subsequently, the samples were divided into training and test sets in an 8:2 ratio. Optimal parameters were selected on the training set using 10-fold cross-validation and grid search, and then validated on the test set. For models, logistic regression, XGBoost, and random forest were considered, and the best-performing model was selected for subsequent analysis. Model construction was completed using Python and related packages.

[0080] In addition, the model was further optimized and evaluated using the following methods: (1) Hyperparameter tuning via grid search; (2) Evaluate the model's discriminative power using ROC curves; (3) Evaluate the consistency between the model's predicted probability and the actual outcome through calibration curves; (4) Evaluate the clinical net benefit of the model through decision curve analysis; (5) Test the robustness of the model by bootstrap resampling or cross-validation.

[0081] Specifically, feature extraction is performed on the standardized clinical dataset to output the baseline clinical features and treatment regimen features for each patient. Specifically, baseline hepatitis B surface antigen, age, hepatitis B virus DNA load, baseline hepatitis B core antigen, neutrophil count, and alanine aminotransferase (ALT) are extracted from the standardized clinical dataset as baseline clinical features. Simultaneously, corresponding treatment regimen-related information is extracted as treatment regimen features. The baseline clinical features, treatment regimen features, and individualized HBsAg decline slopes at multiple treatment stages are then matched, merged, and standardized according to the patient's unique identifier to form a model input feature matrix with one record per patient. The patient's actual response label is used as the training label to construct the modeling feature data. The modeling feature dataset is randomly divided into training and testing sets in an 8:2 ratio. For pre-defined machine learning models, including logistic regression, XGBoost, and random forest models, hyperparameter optimization and model training are performed on the training set using 10-fold cross-validation and grid search, resulting in multiple candidate prediction models. These candidate models are then validated on the testing set. ROC curves are used to evaluate model discrimination, calibration curves to evaluate the consistency between predicted probabilities and actual outcomes, and decision curve analysis to evaluate the model's net clinical benefit. Bootstrap resampling or cross-validation is used to test model robustness, and the best-performing model is selected, resulting in the optimal treatment response prediction model.

[0082] Among them, baseline clinical features are basic clinical information collected from patients before treatment, extracted from the regularized clinical dataset. They are the static input features for the model construction of this invention, specifically including clinical variables that reflect the initial state of the patient's treatment, such as baseline hepatitis B surface antigen level, patient age, hepatitis B virus DNA load, baseline hepatitis B core antigen level, neutrophil count, and alanine aminotransferase level.

[0083] Treatment protocol features are information related to the treatment interventions received by patients, extracted from well-organized clinical datasets. They are one of the input features for the model construction of this invention, specifically including relevant variables reflecting the treatment intervention, such as the type of treatment drug used by the patient, the type of treatment protocol, the dosage, the frequency of administration, and the time of treatment initiation.

[0084] The modeling feature dataset is a feature matrix formed by matching, merging, and standardizing the baseline clinical features, treatment plan features, and individualized HBsAg decline slopes of each patient according to their unique patient ID. It is then integrated with the patient's actual response label as the training label to create the dataset used for model training and validation.

[0085] Step 104: Determine the prediction threshold based on the slope of the total HBsAg decrease for each patient during the target treatment phase and the patient's actual response label, and obtain the optimal response prediction threshold and the high-sensitivity warning threshold.

[0086] It should be noted that the slope of the total HBsAg decrease during the target treatment phase for each patient is paired with the corresponding actual response label of the patient. All possible slope values ​​are traversed as candidate cutoff points. The sensitivity, specificity, and Youden index corresponding to each candidate cutoff point are calculated. The optimal threshold for response prediction is obtained by maximizing the Youden index. Then, the candidate cutoff points are initially screened according to the preset sensitivity. The cutoff point with the highest specificity is selected from the initial screening results to obtain the high-sensitivity warning threshold.

[0087] Furthermore, step 104 may include the following sub-steps: S41. Calculate the performance indicators of candidate cutoff points based on the total HBsAg decrease slope and the actual response label of each patient in the target treatment phase, and output multiple candidate cutoff point performance indicators. S42. Select the optimal cutoff point from multiple candidate cutoff point performance indicators and output the best threshold for response prediction. S43. Perform preliminary screening on the performance indicators of multiple candidate cutoff points according to the preset sensitivity, and then select the cutoff point with the highest specificity from the preliminary screening results and output the high sensitivity warning threshold.

[0088] It should be noted that ROC analysis is performed on the prediction metrics or prediction probabilities to calculate the sensitivity and specificity at different candidate cut-off points, and the optimal cut-off point is determined by maximizing the Youden index. The Youden index (i.e., the candidate cut-off point performance metric) is defined as follows: Youden=Sensitivity+Specificity-1; In the formula, Youden is the Youden index, which is obtained by summing the sensitivity and specificity of the corresponding cutoff point and subtracting 1. The cutoff point corresponding to the maximum value of the index is the basis for selecting the optimal threshold for response prediction. Sensitivity is the percentage of subjects who actually achieve a treatment response and are correctly identified as the responding population at a single candidate cutoff point, with the descent slope of the target treatment stage for each patient as the criterion. Specificity is the percentage of subjects who do not achieve a treatment response and are correctly identified as the non-responding population at a single candidate cutoff point, with the descent slope of the target treatment stage for each patient as the criterion.

[0089] Furthermore, this invention sets a high-sensitivity threshold to minimize missed detections in early warning and screening scenarios. The high-sensitivity threshold is determined as follows: among all candidate cutoff points, cutoff points with a sensitivity not lower than a preset value (preferably 0.85) are first selected, and then the one with the highest specificity is chosen as the high-sensitivity threshold. Therefore, the high-sensitivity threshold is preferentially used in early screening and warning scenarios to minimize the omission of potential respondents.

[0090] It is worth noting that the optimal cutoff value determined by maximizing the Youden index in this invention is -0.034100 log10 IU / mL / week, which is equivalent to a cumulative decrease of approximately 0.82 log10 IU / mL in HBsAg over the first 24 weeks of treatment. Therefore, if the total HBsAg decrease slope from 0 to 24 weeks is ≤ -0.034100 log10 IU / mL / week, it indicates a rapid early HBsAg decrease, with a higher probability of achieving a decrease of ≥1 log10 IU / mL in HBsAg by week 48; if the decrease slope is > -0.034100 log10 IU / mL / week, it indicates a higher risk of subsequent low response.

[0091] In this embodiment, performance metrics for candidate cutoff points are calculated based on the total HBsAg decrease slope and the patient's actual response label during the target treatment phase for each patient. Multiple candidate cutoff point performance metrics are output. Specifically, ROC analysis is performed on the total HBsAg decrease slope and the patient's actual response label during the target treatment phase, traversing all possible slope values ​​as candidate cutoff points. Sensitivity and specificity are calculated for each candidate cutoff point, and the Youden index for each candidate cutoff point is calculated using the formula Youden = Sensitivity + Specificity - 1. This outputs multiple candidate cutoff point performance metrics that include sensitivity, specificity, and the Youden index. Optimal cutoff is then performed on the multiple candidate cutoff point performance metrics. The process involves point screening to output the optimal threshold for response prediction. Specifically, the maximum value of the Youden index is used as the screening criterion to select the corresponding candidate cutoff point as the optimal threshold for response prediction. In this embodiment, the optimal threshold is -0.034100log10 IU / mL / week, which is equivalent to a cumulative decrease of approximately 0.82log10 IU / mL in HBsAg over the first 24 weeks of treatment. Multiple candidate cutoff points are initially screened according to a preset sensitivity. Then, the cutoff point with the highest specificity is selected from the initial screening results to output a high-sensitivity warning threshold. Specifically, candidate cutoff points are initially screened according to a preset sensitivity (preferably 0.85), retaining cutoff points with a sensitivity not lower than the preset value. The cutoff point with the highest specificity is then selected from the initial screening results as the high-sensitivity warning threshold. Sensitivity is defined as the slope of the total HBsAg decrease during the target treatment phase for each patient. At a single candidate cutoff point, the percentage of subjects who correctly identify as responders is the actual treatment response population, and this percentage is used to evaluate the performance of the candidate cutoff point in identifying the responder population. Specificity is defined by the slope of the total HBsAg decrease during the target treatment phase for each patient. It is the percentage of subjects who do not achieve a treatment response at a single candidate cutoff point that are correctly identified as non-responders. It is a performance indicator used to evaluate the ability of candidate cutoff points to identify non-responders.

[0092] Step 105: When the medical data to be tested is received, the optimal treatment response prediction model is used to predict the treatment response probability of each patient in the future treatment stage and compare it with the preset probability threshold to determine the patient's treatment response probability. The result of the determination is then output.

[0093] It should be noted that when the medical data to be tested is received, the data is first preprocessed to obtain a regular clinical dataset. The baseline clinical features and treatment plan features of each patient are extracted, and the individualized HBsAg decline slope for multiple treatment stages is calculated. After the above features are matched and integrated, they are input into the optimal treatment response prediction model. The model makes predictions based on the input features and outputs the treatment response probability of each patient in the future treatment stage. Then, the treatment response probability of each patient is compared with the preset probability threshold one by one, and the treatment response probability judgment result of the patient is output.

[0094] Furthermore, step 105 may include the following sub-steps: S51. Preprocess the medical data to be tested to obtain a well-organized clinical dataset. S52. Perform individualized segmented HBsAg decline slope fitting calculation on the normalized clinical dataset to be tested, and output the individualized HBsAg decline slope of each patient in multiple treatment stages. S53. Extract features from the normalized clinical dataset to be tested, and output the baseline clinical features and treatment plan features of each patient to be tested. S54. Input the baseline clinical characteristics to be tested, the characteristics of the treatment plan to be tested, and the individualized HBsAg decline slope to be tested in multiple treatment stages of each patient into the optimal treatment response prediction model for prediction, and output the treatment response probability of each patient in the future treatment stage. S55. Determine the probability of treatment response for each patient in the future treatment stage against the preset probability threshold, and output the patient's treatment response probability determination result.

[0095] The patient visit data to be tested is newly received patient visit data that has not yet been processed by the process of this invention. It includes the patient's baseline information, treatment information, hepatitis B surface antigen test information, etc., providing the original data basis for subsequent treatment response prediction.

[0096] The standardized clinical dataset to be tested is a standardized dataset obtained after preprocessing the patient visit data. Through variable unification, outlier correction, and missing value imputation, interference factors in the data are eliminated, ensuring the data validity of subsequent feature extraction and model prediction.

[0097] It should be noted that the preprocessing of the patient visit data to be tested yields a well-defined clinical dataset. Specifically, following the same processing rules as the model training phase, the patient visit data undergoes variable format standardization, outlier removal and correction, and missing value imputation. Furthermore, the HBsAg detection data undergoes a base-10 logarithmic transformation to generate a well-defined clinical dataset that perfectly matches the format and specifications of the training phase. Finally, the individualized segmented HBsAg descent slope is fitted and calculated on the well-defined clinical dataset to be tested, outputting the individualized HBsAg values ​​for each patient across multiple treatment stages. The HBsAg decline slope is calculated by using the same piecewise linear mixed-effects model as in the training phase. Based on the HBsAg time-series detection data of the patients before treatment (24 weeks prior to treatment), individualized decline rates are fitted for the 0-12 week and 12-24 week treatment phases, i.e., the individualized HBsAg decline slope. Feature extraction is performed on the regularized clinical dataset to be tested, outputting the baseline clinical features and treatment regimen features for each patient. Specifically, baseline clinical information and treatment regimen information of the same type and dimensions as in the training phase are extracted, including baseline HBsAg levels. The model inputs include patient age, hepatitis B virus DNA load, neutrophil count, alanine aminotransferase (ALT) level, treatment type, and medication regimen, ensuring consistency between the feature inputs and the model training phase. The baseline clinical characteristics, treatment regimen characteristics, and individualized HBsAg decline slopes at multiple treatment stages are input into the optimal treatment response prediction model for each patient. The model outputs the treatment response probability for each patient in the future treatment stage. Specifically, the aforementioned features are processed according to the standardized rules of the training phase and input into the trained optimal model in a single-patient-per-line format. Based on the learned correlation between features and response outcomes, the model outputs the probability value of each patient achieving the preset treatment response outcome in the future treatment stage. The treatment response probability of each patient in the future treatment stage is then compared with a preset probability threshold, and the patient's treatment response probability determination result is output. Specifically, the treatment response probability of each patient is compared with the preset probability threshold; if the probability is not lower than the preset threshold, it is considered a high probability of response; if it is lower, it is considered a low probability of response, generating a clear patient treatment response probability determination result.

[0098] Step 106: Based on the slope of the decrease in total HBsAg, the optimal threshold for response prediction, and the high-sensitivity warning threshold for each patient in the target treatment stage corresponding to the medical data to be tested, perform patient treatment response risk stratification and early warning determination to obtain patient-specific risk stratification early warning results.

[0099] It should be noted that the slope of the decrease in the total HBsAg to be measured for each patient is compared with the optimal threshold for response prediction one by one. Patients who reach or fall below the threshold are classified into the high probability response group, and patients who do not reach the threshold are classified into the high risk non-response group. Then, the slope of the decrease in the total HBsAg to be measured for each patient is compared with the high sensitivity warning threshold, and patients who reach the threshold are marked. Finally, the grouping results and the marking information are integrated to obtain the patient's personalized risk stratification early warning results.

[0100] Furthermore, step 106 may include the following sub-steps: S61. Compare the slope of the decrease in total HBsAg during the target treatment phase of each patient with the optimal threshold for response prediction. Patients whose slope of the decrease in total HBsAg reaches or falls below the optimal threshold for response prediction are classified into the high-probability response group. Patients whose slope of the decrease in total HBsAg does not reach the optimal threshold for response prediction are classified into the high-risk non-response group. Output the initial grouping results. S62. Compare and determine the slope of the decrease in total HBsAg during the target treatment phase of each patient with the high-sensitivity warning threshold, mark the patients whose slope of the decrease in total HBsAg reaches the high-sensitivity warning threshold, and output the early warning marking results. S63. Integrate the initial grouping results and early warning labeling results to output personalized risk stratification early warning results for patients.

[0101] The high-probability response group is a patient group that is divided by comparing the slope of the decrease in total HBsAg to be measured with the optimal threshold for response prediction. Patients in this group are more likely to achieve an effective treatment response in subsequent treatment.

[0102] The high-risk non-response group is a patient group identified by comparing the slope of the decrease in total HBsAg to be measured with the optimal threshold for response prediction. Patients in this group have a higher risk of treatment non-response during subsequent treatment.

[0103] It should be noted that the decline rate of the total HBsAg to be measured in each patient during the target treatment phase is compared with the optimal response prediction threshold for each patient. Specifically, the decline rate of the total HBsAg to be measured for each patient is retrieved sequentially according to the patient's unique number, and the values ​​are compared one by one. Based on preset classification rules, the population is classified. Patients whose decline rate of the total HBsAg to be measured reaches or falls below the optimal response prediction threshold are assigned to the high-probability response group, while patients whose decline rate of the total HBsAg to be measured does not reach the optimal response prediction threshold are assigned to the high-risk non-response group. The grouping information for each patient is recorded simultaneously, and the initial grouping results are output. Then, the decline rate of the total HBsAg to be measured in each patient during the target treatment phase is compared with the high-sensitivity warning threshold. The process involves comparing and judging each case individually, and then screening all the data to be tested according to the patient's number. Patients whose total HBsAg decrease rate reaches a high-sensitivity warning threshold are given a unique warning marker. After recording the marking status of all patients, the early warning marking results are output. Finally, using the patient's unique number as the association basis, the grouping information and warning marker information corresponding to the same patient are summarized and integrated to form a complete data record, and then the patient's personalized risk stratification early warning results are output. This judgment method relies on core indicators and predetermined thresholds in the early stages of treatment to complete automated classification and marking, enabling rapid patient risk identification in the early stages of treatment. This effectively solves the problem that existing methods for predicting the response to treatment of chronic hepatitis B patients cannot meet the needs of early clinical warning.

[0104] Step 107: Integrate the patient treatment response probability determination results and the patient personalized risk stratification early warning results, and output the patient treatment response prediction results.

[0105] It should be noted that the integration of patient treatment response probability assessment results and patient-specific risk stratification early warning results is achieved by using the patient's unique ID as the association index. The two types of result data for each patient are matched and integrated sequentially. The high and low response probability assessment information predicted by the model is merged with the information on high-probability response groups, high-risk non-response groups, and whether an early warning marker is present, all based on slope thresholds, into the same patient record. In practical applications, this can result in eight complete combinations: high response probability combined with a high-probability response group without triggering an early warning marker; high response probability combined with a high-probability response group with an early warning marker; high response probability combined with a high-risk non-response group without triggering an early warning marker; high response probability combined with a high-risk non-response group with an early warning marker; low response probability combined with a high-probability response group without triggering an early warning marker; low response probability combined with a high-probability response group with an early warning marker; low response probability combined with a high-risk non-response group without triggering an early warning marker; and low response probability combined with a high-risk non-response group with an early warning marker. After completing the matching and combination of various information types, the data format is standardized, and the final output is a complete patient treatment response prediction result.

[0106] It is worth mentioning that the patient treatment response prediction results generated by this invention are only used as reference data for clinical efficacy prediction and disease risk screening, and are only used for data statistics and clinical auxiliary reference. They are not used as the basis for prescribing medication plans or implementing clinical diagnosis and treatment, and cannot be directly used to carry out any disease treatment operations for patients with chronic hepatitis B.

[0107] For comparison of technological effectiveness, existing technologies can be used as a reference. Chronic hepatitis B virus (HBV) infection remains a significant global public health issue. According to the World Health Organization, there were approximately 254 million people living with chronic HBV globally in 2022, with about 1.2 million new infections each year. The sustainable development of chronic HBV infection leads to liver fibrosis, cirrhosis, hepatocellular carcinoma (HCC), and related premature deaths, placing a heavy burden on public health systems and clinical management.

[0108] Although the dynamic changes of HBsAg have received widespread attention, existing studies focus more on baseline HBsAg levels or predicting the absolute decrease in HBsAg at the endpoint. There is still a lack of unified, clear and easy-to-implement technical solutions on how to use the early decline trajectory of HBsAg to predict whether HBsAg will decrease by ≥1log10 IU / mL at 48 weeks, thereby achieving early warning and prediction of the disease.

[0109] In existing technologies, some studies have attempted to use machine learning methods to predict HBsAg clearance in CHB patients. Tian et al. conducted a study in 2019, including approximately 2000 patients with chronic hepatitis B from January 2006 to June 2015, and constructed HBsAg clearance prediction models using classification algorithms such as logistic regression, decision trees, random forests, and extreme gradient boosting (XGBoost). The results showed that the XGBoost model achieved an AUC of 0.891, indicating that machine learning methods have high accuracy in predicting HBsAg clearance. However, its main objective is the overall prediction of HBsAg clearance, and it has not yet focused on the interpretable threshold identification and risk stratification of the HBsAg decline rate.

[0110] In recent years, dynamic prediction methods based on longitudinal qHBsAg measurements have been further developed. A prospective follow-up cohort study from multiple centers in China included 6792 CHB patients receiving NAs treatment. Using qHBsAg measurements obtained from multiple follow-ups, longitudinal discriminant analysis was employed to integrate all qHBsAg information during the patient's follow-up period, establishing the GOLDEN model (a predictive model for clinical cure of chronic hepatitis B HBsAg clearance based on the dynamic trajectory of longitudinal quantitative hepatitis B surface antigen (qHBsAg) throughout the treatment course) to predict HBsAg clearance. This model demonstrated extremely high predictive performance with AUCs of 0.981 and 0.979 on the training and external validation sets, respectively. This study suggests that only longitudinal HBsAg trajectories can accurately estimate the probability of HBsAg clearance and has the potential for functional cure assessment and population stratification.

[0111] However, the above technical solutions mainly rely on the integration of the entire longitudinal trajectory and complex discrimination models, and do not explicitly propose: (1) The key treatment time window with the most early warning value in the treatment of hepatitis B; (2) How to use the HBsAg descent rate, a simpler and more easily explained dynamic feature, as the core parameter; (3) How to construct the best cut-off point and risk stratification rules that can be directly used for clinical implementation.

[0112] Therefore, although existing technologies can achieve high-precision prediction, they still have shortcomings in terms of clinical simplicity, interpretability, and early warning applications.

[0113] In summary, existing technologies have provided the following types of solutions: 1. A model for predicting HBsAg clearance based on machine learning algorithms; 2. A high-precision dynamic prediction model based on the entire longitudinal qHBsAg trajectory; 3. A study evaluating the efficacy of HBsAg reduction ≥1 log10 IU / mL as the interim response endpoint at week 48; However, at least the following shortcomings still exist: 1. Current technologies have not yet clearly defined the key time window with the most predictive value in the treatment of hepatitis B, and in particular, there is a lack of systematic refinement of the simple dynamic indicator of the total HBsAg decline slope from 0 to 24 weeks.

[0114] 2. Existing technologies focus more on the level value at a single time point, the entire complex trajectory, or whether the endpoint has occurred, while lacking a prediction scheme based on the interpretable dynamic parameter of HBsAg descent rate.

[0115] Current technology has not yet established a method to predict the HBsAg decline slope based on the most predictive key time window for treatment response at week 48, especially the optimal threshold definition and risk stratification rules for predicting whether an HBsAg decline of ≥1 log10 IU / mL is achieved at week 48.

[0116] As can be seen from the above, the technical problem to be solved by the present invention includes: (1) Identify the treatment period with the most early warning value in the treatment of hepatitis B; (2) Calculate the personalized rate of decrease of hepatitis B surface antigen; (3) Provide a threshold for the rate of decrease to identify high-probability responders and high-risk low-responders at 48 weeks.

[0117] To address the aforementioned issues, this invention provides a method for predicting treatment response in patients with chronic hepatitis B (CHB). Specifically, it is a method for predicting treatment response at week 48 based on the dynamic decline slope of hepatitis B surface antigen (HBsAg) in the 24 weeks prior to treatment. This technical solution collects, processes, models, and determines thresholds based on longitudinal changes in HBsAg in patients over the 24 weeks prior to treatment, enabling early prediction and risk stratification of subsequent treatment outcomes, thereby providing support for clinical treatment decisions. Specifically, it can identify the critical time window for CHB treatment and predict whether patients will reach a preset response endpoint (e.g., a response or HBsAg decrease of ≥1 log10 IU / mL at week 48) based on key dynamic changes in HBsAg before treatment. Furthermore, it provides clear thresholds and risk stratification rules to meet the needs of early clinical warning, efficacy stratification, and treatment decision support.

[0118] Specifically, such as Figure 2As shown, the study first included enrolling patients with chronic hepatitis B who met the criteria for interferon treatment between May 2018 and December 2025. This formed the study population. Subsequently, comprehensive data collection was conducted on the enrolled patients, including gender, age, lifestyle, treatment regimen, past medication history, family history, and laboratory test results for hepatitis B surface antigen, hepatitis B core antigen, hepatitis B DNA viral load, alanine aminotransferase (ALT), and neutrophil count. The raw data were then preprocessed, including variable name and unit of measurement standardization, outlier correction, and missing value imputation. Simultaneously, a base-10 logarithmic transformation was performed on the HBsAg index. After preprocessing, an auxiliary analysis module integrating Kaplan-Meier curves, log-rank tests, linear regression, logistic regression, and Markov chain analysis was used to perform data analysis. Next, dynamic feature construction was performed, building a piecewise linear mixed-effects model based on three treatment intervals: 0-12 weeks, 12-24 weeks, and 24-36 weeks. Based on this model, slope_mixed_0_12 and slope_mixed_12_24, and the individualized decline slope in each stage were extracted. After integration and calculation, the total decline slope slope_mixed_0_24 from week 0 to 24 was obtained. The baseline variable and the HBsAg decline rate in each stage were used as input features for machine learning model construction. During the modeling stage, the training set and test set were split in an 8:2 ratio. Five-fold cross-validation and grid search were used to optimize the model hyperparameters. After the model was built, ROC curves, calibration curves, DCA analysis, and bootstrap resampling were used to conduct a comprehensive model evaluation to select the optimal prediction model. Based on the optimal model, risk stratification and result output were carried out. The subjects were divided into a high-probability response group and a high-risk group. The model simultaneously predicted whether the patients could become responders in week 48 and whether they could achieve the response standard of HBsAg decline ≥1log10 IU / mL. Finally, the threshold was defined based on the Youden index, and the optimal threshold for response prediction and the high-sensitivity warning threshold were obtained respectively. Subsequently, two thresholds were used to compare the total decline slope of all subjects from week 0 to 24, and the patients were grouped and classified and potential responders were marked in sequence. The results of individualized risk stratification of patients were then generated. The results of the early warning were then input into the optimal machine learning model selected by screening for batch calculation, and finally the treatment response prediction results for each patient were output. The prediction results are only used as reference information for clinical efficacy prediction and population risk screening, and cannot be used as the basis for formulating medication plans and carrying out clinical treatment interventions.

[0119] In summary, the key points of this invention are: 1) A technical solution based on longitudinal HBsAg data from 36 weeks before treatment to extract individualized segmented slopes for weeks 0–12, 12–24, and 24–36.

[0120] 2) Based on the optimal prediction window of 0–24 weeks, the total slope of 0–24 weeks is calculated to predict the treatment response at 48 weeks, and the optimal cut-off point of the total slope of 0–24 weeks is determined by ROC / Youden.

[0121] 3) Risk stratification rule with slope_mixed_0_24 ≤ -0.034393 log10 IU / mL / week as the threshold for high probability response.

[0122] In addition to 48-week treatment response, the outcome can be replaced with HBsAg clearance, functional cure, virological response, or other clinical endpoints.

[0123] Compared with the prior art, the present invention has the following advantages: First, it fully integrates the segmented dynamic features of the 24 weeks before treatment in the treatment process rather than relying on a single time point or a single time window; Second, it achieves high predictive performance while maintaining model interpretability, with an AUC of 0.942 in this embodiment; Third, it provides a clear and operable total slope cut-off point, enabling the model to be directly used for clinical screening, early warning, and patient stratification.

[0124] It is worth mentioning that, through statistical analysis, this invention identified 1,244 HBV patients initially, of whom 374 were excluded due to insufficient follow-up (n=251), cirrhosis / hepatocellular carcinoma / hepatitis C virus infection (n=89), age under 18 years or over 65 years (n=26), or pregnancy (n=8). Of the remaining 870 patients followed up to week 12, 357 had treatment durations of less than 48 weeks, and ultimately 513 patients were included in the week 48 analysis.

[0125] Variables with missing values ​​>20% were excluded. Continuous variables were outlier detected using the 1.5-fold interquartile range (IQR) rule and truncated by 1% above and below the threshold for winsorization. Missing values ​​were imputed using the missForest method. Patient data were divided into continuous and categorical variables. Normality of continuous variables was assessed using the Kolmogorov-Smirnov test. Normally distributed variables were described as mean ± standard deviation, and skewed variables as median and interquartile range. Categorical variables were reported as number of cases and percentage. Since HBsAg decline during treatment is a dynamic process, this study used a discrete-state Markov model to describe the longitudinal transitions between clinically significant HBsAg decline states. This method estimates the conditional probability of transitioning from one decline state to another within a continuous treatment interval and identifies early states associated with persistently low response or subsequent improvement. Data analysis was performed using SPSS 29.0 (IBM Corp., Armonk, NY, USA), Python (version 3.10.19), and R (version 4.5.2).

[0126] A total of 513 CHB patients were included in the analysis, of whom 264 achieved a ≥1 log decrease in HBsAg after 48 weeks of treatment, and 249 achieved a <1 log decrease. The overall response rate at week 48 was 51.5%. The mean age at baseline was 37.9 years (range 18–65 years; SD 8.74). Baseline characteristics stratified by response status at week 48 are shown in Table 1. Compared with patients whose HBsAg decreased by <1 log at week 48, patients whose HBsAg decreased by ≥1 log were younger, had a lower proportion of prior IFN treatment, and were more likely to be treatment-first. The HBsAg level at week 12 was significantly lower in the ≥1 log decrease group than in the <1 log decrease group. No statistically significant differences were observed between the two groups in terms of sex, BMI, baseline HBsAg, baseline HBeAg, baseline HBV DNA, baseline neutrophil count, or baseline ALT.

[0127] Table 1. Baseline characteristics of HBV participants and their association with HBsAg decline at 48 weeks of treatment.

[0128] Overall, HBsAg levels gradually decreased during the 48-week treatment period, with the most significant decreases occurring in the early and middle stages of treatment. Different HBsAg dynamics were observed after stratification based on whether a ≥1 log decrease was achieved at week 48. Patients with a ≥1 log decrease in HBsAg at week 48 exhibited significant and sustained HBsAg level declines, particularly during weeks 12–24 of treatment, followed by further reductions at week 48. In contrast, patients with a <1 log decrease in HBsAg at week 48 maintained relatively higher HBsAg levels throughout the follow-up period with smaller fluctuations over time.

[0129] Discrete-state Markov models were used to describe longitudinal metastasis of HBsAg decline relative to baseline during treatment (Table 2). Results suggest that early HBsAg decline status has clear early warning value for subsequent treatment response. Patients were categorized into four statuses based on the magnitude of HBsAg decline: S0 (decline <0.5 log10); S1 (decline 0.5–1 log10); S2 (decline 1–2 log10); and S3 (decline ≥2 log10). At week 12, most patients remained in a low-decline status, with 57.6% in S0, 17.9% in S1, and only 16.8% and 7.7% reaching S2 and S3, respectively. Between weeks 12 and 24, the probability of metastasis differed significantly depending on the decline status at week 12. Among patients classified as S1 at week 12, 46.0% metastasized to S2 and 17.7% to S3 by week 24. In contrast, among patients still in S0 at week 12, only 11.3% progressed to S2 and 2.8% to S3 by week 24. These results suggest that early HBsAg decline has early warning value for subsequent treatment response. The cumulative responder proportion observed within 48 weeks was highly close to the Markov predicted proportion, with a mean absolute error of 0.0220 and a root mean square error of 0.0265. By combining fixed effects and individual-specific random effects, individualized slopes for the week 0–12 and week 12–24 intervals were obtained and used as dynamic predictors in subsequent analyses.

[0130] Table 2. Probability of HBsAg Decline in Different Treatment Intervals

[0131] The final cohort was randomly divided into a training set and a test set using stratified sampling at a ratio of 7:3. The training set included 359 patients, and the test set included 154 patients. Baseline characteristics and early treatment indicators were comparable between the two groups. Comparison of baseline and 24-week treatment characteristics between the training and validation sets (Table 3) showed no significant differences in any indicators.

[0132] Table 3. Comparison of baseline features between training and test sets

[0133] Furthermore, as shown in Table 4, this invention uses univariate logistic regression to assess the association between each candidate predictor and HBsAg response at week 48. Variables with clinical relevance and statistical evidence of association are included in multivariate modeling.

[0134] Multivariate regression analysis showed that prior IFN treatment history, the HBsAg decline slope at weeks 0-12, and the HBsAg decline slope at weeks 12-24 were independent predictors of HBsAg response at week 48, all with statistical significance (P < 0.001). Other variables (age, sex, BMI, baseline HBsAg level, HBV DNA load, baseline ALT, baseline neutrophil count, etc.) did not show independent statistical association in the multivariate model (P > 0.05), and their potential impact was adjusted for by other significant variables in the model. Specifically, for patients with a prior IFN treatment history, the odds ratio (OR) for HBsAg response at week 48 was 3.07 (95% confidence interval (CI)). Interval: 1.65-5.74, P<0.001), indicating that a history of IFN treatment was independently associated with a higher probability of response; for every additional 0.1 log10 IU / mL decrease in HBsAg level during weeks 0-12, the log odds of HBsAg response at week 48 was significantly increased, with an OR of 0.86 (95% CI: 0.83-0.89, P<0.001), reflecting an independent positive correlation between the rapid decrease in HBsAg in the early stages of treatment and the increased probability of response; for every additional 0.1 log10 IU / mL decrease in HBsAg level during weeks 12-24, the log odds of HBsAg response at week 48 was also significantly increased, with an OR of 0.90 (95% CI: 0.86-0.93, P<0.001), indicating that the sustained decrease in HBsAg during the middle stages of treatment was an independent protective factor for response. In summary, multivariate logistic regression analysis confirmed that, after controlling for confounding factors, the history of previous IFN treatment and the slope of dynamic HBsAg decline in the early and middle stages of treatment were the core independent predictors of HBsAg response at week 48. Among them, the independent predictive value of dynamic HBsAg change indicators was particularly outstanding, which can provide reliable evidence-based basis for individualized treatment response risk stratification and early intervention decisions for patients with chronic hepatitis B.

[0135] Table 4. Univariate and multivariate logistic regression analyses of factors related to HBsAg response in the training set at week 48.

[0136] Furthermore, feature selection was performed on the training set. After LASSO regression (Least Absolute Shrinkage and Selection Operator, L1 regularized regression algorithm), collinearity assessment, and logistic regression screening (Table 3), 10 predictors were retained for model development: age, BMI, IFN treatment history, baseline HBsAg, HBeAg, HBV DNA, neutrophils, HBsAg slope at weeks 0-12, and HBsAg slope at weeks 12-24. Six candidate models were constructed on the training set and evaluated on a test set of 154 cases, including LR (Logistic Regression), DT (Decision Tree), RF (Random Forest), XGBoost (Extreme Gradient Boosting), SVM (Support Vector Machine), and ANN (Artificial Neural Network). All candidate models demonstrated good discriminative power in predicting a significant decrease in HBsAg at week 48, with AUCs ranging from 0.876 to 0.905. RF showed the highest numerical discriminative power (AUC: 0.905, 95% CI 0.854–0.949) (Table 5). At a classification threshold of 0.5, RF achieved an accuracy of 0.838, precision of 0.847, sensitivity of 0.813, specificity of 0.861, and an F1 score of 0.830. This model correctly classified 61 out of 75 responders and 68 out of 79 non-responders. RF had the lowest Brier score but performed well in DCA. Overall, RF achieved the best balance in terms of discriminative power, calibration, clinical usability, and interpretability, and was therefore selected as the final predictive model.

[0137] Table 5 Comparison of candidate models in the test set

[0138] Furthermore, the HBsAg slope from week 0 to 24 was evaluated as a clinically interpretable early risk stratification biomarker. The HBsAg slope from week 0 to 24 showed strong discriminatory power against non-response at week 48, with an AUC of 0.9109. The optimal cutoff value determined based on the maximum Youden index was -0.034100log10 IU / mL / week, with a sensitivity of 86.35%, specificity of 84.47%, and a Youden index of 0.7082. This threshold corresponds to a cumulative HBsAg decrease of approximately 0.82log10 IU / mL at week 24. A high-sensitivity warning threshold was also derived based on a preset sensitivity target of at least 0.85, with a selected warning cutoff value of -0.033764log10 IU / mL / week, exhibiting a sensitivity of 85.54% and specificity of 84.85%.

[0139] When using the Youden cutoff value, patients with an HBsAg slope greater than -0.034100log10 IU / mL / week from week 0 to 24 were classified as high-risk non-responders, while patients with a slope equal to or below the threshold were classified as likely responders.

[0140] In this embodiment of the invention, a method for predicting treatment response in patients with chronic hepatitis B is provided. The method involves acquiring raw medical data and preprocessing it to obtain a regularized clinical dataset. The dynamic slope of HBsAg is calculated on the regularized clinical dataset to obtain the individualized HBsAg decline slope for each patient across multiple treatment stages and the total HBsAg decline slope for each patient in the target treatment stage. Based on the patient's actual response label, the individualized HBsAg decline slope for each patient across multiple treatment stages, and the regularized clinical dataset, a model is constructed and optimized to obtain an optimal treatment response prediction model. A prediction threshold is defined based on the total HBsAg decline slope for each patient in the target treatment stage and the patient's actual response label to obtain an optimal response prediction threshold and a high-sensitivity warning threshold. When medical data to be tested is received, the optimal treatment response prediction model is used to predict the response based on the medical data, outputting the treatment response probability for each patient in the future treatment stage and comparing it with a preset probability threshold to determine the patient's treatment response probability. The result of the patient's treatment response probability determination is then output. The method is based on the corresponding medical data to be tested. For each patient in the target treatment phase, the slope of the decrease in total HBsAg, the optimal threshold for response prediction, and the high-sensitivity warning threshold are used to stratify the patient's treatment response risk and determine early warning, resulting in personalized risk stratification and early warning results. The patient's treatment response probability determination results and personalized risk stratification and early warning results are integrated to output the patient's treatment response prediction results. Based on the above, this invention relies on the dynamic HBsAg decrease characteristics in the early stages of treatment to conduct quantitative analysis and model training. This fully explores the efficacy prediction value of early treatment indicators, overcoming the shortcomings of traditional prediction methods, such as strong lag and inability to make early judgments. By establishing a dual-judgment system combining machine learning multi-feature comprehensive prediction and slope threshold stratification and early warning, it not only utilizes the model to integrate multi-dimensional early data to ensure the overall accuracy of response prediction, but also achieves rapid risk stratification and early warning based on the early antigen decrease trend through exclusive quantitative thresholds. This can accurately identify the patient's long-term response risk in the early stages of treatment, enabling early prediction and risk warning of treatment response in patients with chronic hepatitis B, and meeting the needs of early clinical diagnosis and treatment assessment.

[0141] Please see Figure 3 , Figure 3 This is a structural block diagram of a treatment prediction and response system for patients with chronic hepatitis B, provided in Embodiment 2 of the present invention.

[0142] This invention provides a treatment prediction response system for patients with chronic hepatitis B, comprising: The acquisition module 301 is used to acquire raw medical data and preprocess the raw medical data to obtain a regularized clinical dataset. The calculation module 302 is used to calculate the dynamic slope of HBsAg on the regularized clinical dataset to obtain the individualized HBsAg decline slope for each patient in multiple treatment stages and the total HBsAg decline slope for each patient in the target treatment stage. Module 303 is used to build and optimize the model based on the patient's actual response label, the individualized HBsAg decline slope of each patient in multiple treatment stages, and the regularized clinical dataset, so as to obtain the optimal treatment response prediction model. The definition module 304 is used to define the prediction threshold based on the slope of the total HBsAg decrease for each patient in the target treatment phase and the patient's actual response label, so as to obtain the optimal response prediction threshold and the high-sensitivity warning threshold. The prediction module 305 is used to predict the treatment response probability of each patient in the future treatment stage by using the optimal treatment response prediction model when receiving the medical data to be tested, and to judge the patient's treatment response probability one by one with the preset probability threshold. The prediction module 305 is used to predict the treatment response probability of each patient in the future treatment stage by using the optimal treatment response prediction model when receiving the medical data to be tested, and to judge the patient's treatment response probability one by one. The judgment module 306 is used to perform patient treatment response risk stratification and early warning judgment based on the decline slope of the total HBsAg to be tested, the optimal threshold for response prediction, and the high-sensitivity early warning threshold of each patient in the target treatment stage, so as to obtain the patient's personalized risk stratification early warning result. The output module 307 is used to integrate the patient treatment response probability determination results and the patient personalized risk stratification early warning results to output the patient treatment response prediction results.

[0143] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system and modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0144] This invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program; when the computer program is executed by the processor, the processor performs the steps of the treatment prediction response method for patients with chronic hepatitis B as described in the above embodiments.

[0145] This invention also provides a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements the steps of the treatment prediction response method for patients with chronic hepatitis B as described in the above embodiments.

[0146] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0147] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0148] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0149] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0150] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting treatment response in patients with chronic hepatitis B, characterized in that, include: Obtain raw medical records and preprocess them to obtain a regularized clinical dataset. The dynamic slope of HBsAg was calculated on the regularized clinical dataset to obtain the individualized HBsAg decline slope for each patient in multiple treatment stages and the total HBsAg decline slope for each patient in the target treatment stage. The optimal treatment response prediction model is obtained by optimizing the model based on the patient's actual response label, the individualized HBsAg decline slope of each patient in multiple treatment stages, and the regularized clinical dataset. Based on the slope of the total HBsAg decrease for each patient during the target treatment phase and the patient’s actual response label, a prediction threshold is defined to obtain the optimal response prediction threshold and the high-sensitivity warning threshold. When the medical data to be tested is received, the optimal treatment response prediction model is used to predict the treatment response probability of each patient in the future treatment stage and judge it one by one with the preset probability threshold, and output the patient treatment response probability judgment result. Based on the slope of the decrease in total HBsAg in the target treatment stage for each patient corresponding to the medical data to be tested, the optimal threshold for response prediction, and the high-sensitivity early warning threshold, the patient treatment response risk stratification and early warning determination are performed to obtain the patient personalized risk stratification early warning results. By integrating the patient's treatment response probability determination results and the patient's personalized risk stratification early warning results, the patient's treatment response prediction results are output.

2. The method for predicting treatment response in patients with chronic hepatitis B according to claim 1, characterized in that, The preprocessing of the original medical records to obtain a standardized clinical dataset includes: The original medical records are filtered to output a cohort of patients receiving treatment for chronic hepatitis B. Treatment data were collected from the cohort of patients with chronic hepatitis B and the raw clinical dataset was output. The original clinical dataset is processed sequentially with variable unification, outlier correction, logarithmic transformation, and missing value imputation to output a normalized clinical dataset.

3. The method for predicting treatment response in patients with chronic hepatitis B according to claim 1, characterized in that, The step of calculating the dynamic slope of HBsAg on the regularized clinical dataset to obtain the individualized HBsAg decline slope for each patient at multiple treatment stages and the total HBsAg decline slope for each patient at the target treatment stage includes: The individualized segmented HBsAg decline slope is fitted and calculated on the regularized clinical dataset, and the individualized HBsAg decline slope for each patient in multiple treatment stages is output. The individualized HBsAg reduction slopes of each patient at each treatment stage were integrated to obtain the total HBsAg reduction slope for each patient at the target treatment stage.

4. The method for predicting treatment response in patients with chronic hepatitis B according to claim 1, characterized in that, The optimal treatment response prediction model is obtained by optimizing the model based on the patient's actual response label, the individualized HBsAg decline slope for each patient across multiple treatment stages, and the regularized clinical dataset, including: Feature extraction is performed on the regularized clinical dataset to output the baseline clinical characteristics and treatment plan characteristics of each patient. The baseline clinical characteristics, treatment plan characteristics, and individualized HBsAg decline slopes at multiple treatment stages of each patient are matched, merged, and standardized according to the patient's unique ID to form a model input feature matrix with one record per line for each patient. The actual response labels of the patients are used as training labels to construct a modeling feature dataset. The modeling feature dataset is randomly divided to output a model training set and a model test set. The preset machine learning model is subjected to hyperparameter optimization and model training on the model training set, and multiple candidate prediction models are output. Multiple candidate prediction models are validated on the model test set, and the model with the best performance is selected and the optimal treatment response prediction model is output.

5. The method for predicting treatment response in patients with chronic hepatitis B according to claim 1, characterized in that, The step of defining prediction thresholds based on the slope of the total HBsAg decrease for each patient during the target treatment phase and the patient's actual response label, to obtain the optimal response prediction threshold and the high-sensitivity warning threshold, includes: Based on the total HBsAg decrease slope of each patient in the target treatment phase and the actual response label of the patient, the performance index of the candidate cutoff point is calculated, and multiple candidate cutoff point performance indices are output. Optimal interception points are selected from multiple candidate interception point performance metrics, and the optimal threshold for response prediction is output. The candidate cutoff point performance indicators are initially screened according to a preset sensitivity, and then the cutoff point with the highest specificity is selected from the initial screening results to output a high-sensitivity early warning threshold.

6. The method for predicting treatment response in patients with chronic hepatitis B according to claim 1, characterized in that, The optimal treatment response prediction model is used to predict the treatment response probability of each patient in the future treatment stage based on the medical data to be tested. This probability is then compared with a preset probability threshold, and the patient's treatment response probability determination result is output, including: The medical data to be tested is preprocessed to obtain a regularized clinical dataset. The individualized segmented HBsAg decline slope of the measured regular clinical dataset is fitted and calculated, and the individualized HBsAg decline slope of each patient in multiple treatment stages is output. Feature extraction is performed on the regular clinical dataset to be tested, and the baseline clinical features and treatment plan features of each patient are output. The baseline clinical characteristics to be tested, the characteristics of the treatment plan to be tested, and the individualized HBsAg decline slope to be tested in multiple treatment stages of each patient are respectively input into the optimal treatment response prediction model for prediction, and the probability of treatment response for each patient in future treatment stages is output. The probability of each patient's treatment response in the future treatment stage is determined against a preset probability threshold, and the result of the determination of the patient's treatment response probability is output.

7. The method for predicting treatment response in patients with chronic hepatitis B according to claim 1, characterized in that, The process involves stratifying patient treatment response risk and determining early warning based on the slope of the decrease in total HBsAg during the target treatment phase for each patient corresponding to the medical data to be tested, the optimal threshold for response prediction, and the high-sensitivity warning threshold, to obtain personalized risk stratification and early warning results for patients, including: The slope of the decrease in total HBsAg during the target treatment phase for each patient is compared with the optimal threshold for response prediction. Patients whose slope of the decrease in total HBsAg reaches or falls below the optimal threshold for response prediction are classified into the high-probability response group, while patients whose slope of the decrease in total HBsAg does not reach the optimal threshold for response prediction are classified into the high-risk non-response group. The initial grouping results are then output. The slope of the decrease in total HBsAg during the target treatment phase of each patient is compared with the high-sensitivity warning threshold. Patients whose slope of the decrease in total HBsAg reaches the high-sensitivity warning threshold are marked, and the early warning marking results are output. The initial grouping results and the early warning labeling results are integrated to output personalized risk stratification early warning results for patients.

8. A treatment prediction and response system for patients with chronic hepatitis B, characterized in that, include: The acquisition module is used to acquire raw medical data and preprocess the raw medical data to obtain a regularized clinical dataset. The calculation module is used to calculate the dynamic slope of HBsAg on the regularized clinical dataset to obtain the individualized HBsAg decline slope for each patient in multiple treatment stages and the total HBsAg decline slope for each patient in the target treatment stage. The construction module is used to optimize the model based on the patient's actual response label, the individualized HBsAg decline slope of each patient in multiple treatment stages, and the regularized clinical dataset, so as to obtain the optimal treatment response prediction model. The definition module is used to define the prediction threshold based on the slope of the total HBsAg decrease for each patient in the target treatment phase and the patient’s actual response label, so as to obtain the optimal response prediction threshold and the high-sensitivity warning threshold. The prediction module is used to predict the treatment response probability of each patient in the future treatment stage by using the optimal treatment response prediction model when receiving the medical data to be tested, and to judge the patient's treatment response probability one by one with the preset probability threshold. The prediction module is used to output the patient's treatment response probability judgment result. The determination module is used to perform patient treatment response risk stratification and early warning determination based on the decline slope of the total HBsAg to be tested for each patient in the target treatment stage, the optimal threshold for response prediction, and the high-sensitivity early warning threshold, so as to obtain the patient's personalized risk stratification early warning result. The output module is used to integrate the patient treatment response probability determination results and the patient personalized risk stratification early warning results to output the patient treatment response prediction results.

9. An electronic device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the treatment prediction response method for patients with chronic hepatitis B as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the treatment prediction response method for patients with chronic hepatitis B as described in any one of claims 1-7.

Images