Method and system for predicting and monitoring liver injury risk of anti-tuberculosis treatment

Abstract

This invention discloses a method and system for predicting and monitoring the risk of liver injury during tuberculosis treatment, belonging to the field of medical information processing technology. The method includes: collecting patient baseline data and inputting it into a risk prediction model to calculate a baseline risk score, outputting a three-level risk stratification result and monitoring frequency recommendations; collecting dynamic monitoring data according to the monitoring frequency, analyzing the trend and rate of change, and performing Bayesian dynamic risk updates; outputting graded early warning alerts based on the degree of deviation and rate of change of indicators; and generating a risk assessment report and treatment plan adjustment recommendations. This invention achieves closed-loop management from static risk assessment to dynamic risk updates by fusing baseline data and dynamic monitoring data, improving the accuracy and timeliness of liver injury risk prediction.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, specifically to a method and system for predicting and monitoring the risk of liver damage during tuberculosis anti-tuberculosis treatment. Background Technology

[0002] According to existing clinical studies, the incidence of drug-induced liver injury (DILI) varies considerably among different populations, with an overall incidence ranging from approximately 2% to 28%. The clinical manifestations of DILI are broad, ranging from mild cases presenting only as asymptomatic elevated liver enzymes to severe cases progressing to acute liver failure and even death. DILI not only directly endangers patients' lives and health but may also lead to the interruption of anti-tuberculosis treatment, thereby affecting the cure rate of tuberculosis and increasing the risk of drug-resistant tuberculosis.

[0003] In existing technologies, the prevention and management of drug-induced liver injury in tuberculosis mainly rely on regular liver function monitoring and empirical risk assessment. CN118586022A discloses a data encryption processing system and method for patients with chronic hepatitis B. This technical solution uses a data acquisition module to collect patient diagnosis and treatment data, utilizes a gradient boosting decision tree model to assess the security level of the verification data, and dynamically sets the encryption strategy through a multi-objective particle swarm optimization algorithm. However, this technical solution primarily addresses the privacy protection of medical data; its core lies in data encryption and security level classification, but it does not address the risk prediction and monitoring of drug-induced liver injury.

[0004] Existing methods for assessing the risk of anti-tuberculosis drug-induced liver injury have the following technical shortcomings. First, existing methods primarily rely on static baseline characteristics for risk stratification, such as age, baseline liver function indicators, and hepatitis B virus infection status. They fail to fully utilize dynamically acquired monitoring data during treatment to update risk assessment results in real time, leading to a lack of timeliness and accuracy in risk prediction. Second, existing monitoring strategies lack individualization, typically employing fixed monitoring frequencies without dynamically adjusting monitoring intervals based on the patient's actual risk level. This can result in over-monitoring of low-risk patients or under-monitoring of high-risk patients. Third, existing early warning methods mainly focus on whether the absolute values ​​of liver function indicators exceed preset thresholds, neglecting the trend and rate of change in these indicators, making it difficult to promptly identify early signs of liver injury. Fourth, there is a lack of technical solutions to effectively integrate baseline risk assessment with dynamic monitoring data, hindering the achievement of closed-loop management from static risk stratification to dynamic risk updates.

[0005] Therefore, there is an urgent need to develop a method for predicting and monitoring the risk of liver injury during tuberculosis treatment that can integrate patient baseline characteristics with dynamic monitoring data, achieve individualized risk prediction and adaptive monitoring strategy adjustment, so as to improve the early identification rate of drug-induced liver injury, optimize the allocation of monitoring resources, and ensure the safety of anti-tuberculosis treatment. Summary of the Invention

[0006] To address the aforementioned technical problems in the existing technology, this invention provides a method and system for predicting and monitoring the risk of liver damage during tuberculosis anti-tuberculosis treatment.

[0007] According to a first aspect of the present invention, a method for predicting and monitoring the risk of liver injury during tuberculosis anti-tuberculosis treatment is provided, comprising the following steps:

[0008] Step S1: Collect baseline data of tuberculosis patients. Baseline data includes demographic data, basic liver function indicators, viral infection status, lifestyle data, concomitant medication data, and nutritional status data.

[0009] Step S2: Input the baseline data into the liver injury risk prediction model, calculate the baseline risk score, output the three-level risk stratification results of high risk, intermediate risk and low risk based on the baseline risk score, and generate corresponding monitoring frequency recommendations based on the risk stratification results.

[0010] Step S3: During the anti-tuberculosis treatment, collect dynamic monitoring data of liver function indicators according to the recommended monitoring frequency. Analyze the changing trend and rate of change of each indicator in the dynamic monitoring data through the change trend analysis module, and perform Bayesian dynamic update of the baseline risk score based on the dynamic monitoring data.

[0011] Step S4: The early warning judgment module comprehensively judges the early signs of liver damage based on the degree and rate of change of liver function indicators from the baseline, and outputs graded early warning alerts.

[0012] Step S5: The assessment report generation module generates a liver injury risk assessment report and treatment plan adjustment suggestions based on the risk stratification results, dynamic monitoring analysis results, and early warning judgment results.

[0013] According to a second aspect of the present invention, a system for predicting and monitoring the risk of liver injury during tuberculosis anti-tuberculosis treatment is provided, comprising:

[0014] The baseline data acquisition module is used to collect baseline data of tuberculosis patients. The baseline data includes demographic data, basic liver function index data, viral infection status data, lifestyle data, combined medication data, and nutritional status data.

[0015] The risk prediction module is used to input baseline data into the liver injury risk prediction model, calculate the baseline risk score, output the three-level risk stratification results of high risk, intermediate risk and low risk based on the baseline risk score, and generate corresponding monitoring frequency recommendations based on the risk stratification results.

[0016] The dynamic monitoring module is used to collect dynamic monitoring data of liver function indicators according to the recommended monitoring frequency during anti-tuberculosis treatment. The change trend analysis module analyzes the change trend and rate of change of each indicator in the dynamic monitoring data, and performs Bayesian dynamic updates on the baseline risk score based on the dynamic monitoring data.

[0017] The early warning judgment module is used to comprehensively judge the signs of early liver damage based on the degree and rate of change of liver function indicators from the baseline, and output graded early warning alerts.

[0018] The assessment report generation module is used to generate liver injury risk assessment reports and treatment plan adjustment suggestions based on risk stratification results, dynamic monitoring analysis results, and early warning judgment results.

[0019] The present invention has the following beneficial effects:

[0020] First, by constructing a deep fusion architecture of baseline data and dynamic monitoring data, this invention achieves closed-loop management from static risk assessment to dynamic risk updates, overcoming the limitations of existing technologies that rely solely on baseline features for risk prediction, and significantly improving the timeliness and accuracy of risk prediction.

[0021] Second, this invention achieves individualized adjustment of monitoring frequency through an adaptive monitoring strategy driven by risk stratification results. This avoids over-monitoring of low-risk patients while ensuring that high-risk patients receive sufficiently intensive monitoring attention, thus optimizing the allocation of medical resources.

[0022] Third, this invention achieves early identification of early signs of liver injury through comprehensive analysis of the trends and rates of change of liver function indicators. Compared with existing technologies that only focus on the absolute values ​​of indicators, this invention can advance the warning time and gain valuable time for clinical intervention.

[0023] Fourth, this invention achieves effective integration of baseline prior risk and dynamic monitoring evidence through a Bayesian dynamic risk update mechanism, enabling the risk assessment results to be continuously optimized as monitoring data accumulates, thereby improving the adaptive capability of the prediction model. Attached Figure Description

[0024] Figure 1 This is a flowchart of a method for predicting and monitoring the risk of liver damage during tuberculosis treatment, provided in an embodiment of the present invention.

[0025] Figure 2 This is an architecture diagram of the tuberculosis anti-tuberculosis treatment liver injury risk prediction and monitoring system provided in this embodiment of the invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the present invention.

[0027] This invention provides a method for predicting and monitoring the risk of liver injury during anti-tuberculosis treatment, such as... Figure 1 As shown, the method includes steps S1 to S5, forming a deeply coupled closed-loop collaborative architecture. The output of the previous step serves as the key input of the next step, and the results of the subsequent steps can influence the parameter settings of the previous steps, thereby achieving dynamic adaptive optimization of risk prediction and monitoring.

[0028] Step S1: Baseline data acquisition.

[0029] In one embodiment of the present invention, a baseline data acquisition module is used to comprehensively collect the patient's baseline characteristic data before the initiation of anti-tuberculosis treatment. The completeness and accuracy of the baseline data directly affect the assessment accuracy of the subsequent risk prediction model; therefore, the present invention has systematically designed the scope of baseline data collection and data quality control.

[0030] Specifically, the baseline data includes six core data elements. The first category is demographic data, specifically patient age, sex, and body mass index (BMI). Age is calculated in years; studies have shown that patients aged 35 years or older have a significantly increased risk of drug-induced liver injury, and this invention incorporates age as an important risk assessment factor into the predictive model. BMI is calculated using the formula weight divided by the square of height, in kg / m². Low-weight patients with a BMI below 18.5 kg / m² have poorer nutritional status, reduced liver metabolic reserve capacity, and a correspondingly increased risk of liver injury.

[0031] The second category is basic liver function indicators, specifically including baseline values ​​for alanine aminotransferase (ALT), aspartate aminotransferase (AST), total bilirubin, and alkaline phosphatase (ALP). The upper limit of the normal reference range for ALT is typically set at 40 U / L, for AST at 40 U / L, for total bilirubin at 21 μmol / L, and for ALP at 130 U / L. In one embodiment of the invention, even if baseline liver function indicators are within the normal range, their levels are positively correlated with the risk of subsequent liver injury. For example, patients with a baseline ALT greater than or equal to 16 U / L have a higher risk of liver injury compared to patients with lower baseline values.

[0032] The third category is viral infection status data, specifically including hepatitis B surface antigen (HBsAg) test results and hepatitis C antibody (HCV) test results. HBsAg positivity is an independent risk factor for anti-tuberculosis drug-induced liver injury, and the risk prediction model of this invention assigns a higher risk weight to HBsAg-positive patients. Preferably, for HBsAg-positive patients, HBV DNA load data is further collected, with a viral load greater than 10... 4 Patients with active infections at IU / mL have a higher risk.

[0033] The fourth category is lifestyle data, specifically including alcohol consumption history and smoking history. Alcohol consumption history was assessed using a standardized questionnaire, categorizing patients into four groups: never drinker, former alcohol abstinent, current light drinker, and current heavy drinker. Patients with active alcohol consumption, especially heavy drinkers, have a significantly increased risk of drug-induced liver injury and are more likely to progress to severe liver injury with jaundice.

[0034] The fifth category is data on combined medication use, specifically including the types and quantities of hepatotoxic drugs used concurrently. This invention establishes a list of hepatotoxic drugs, including but not limited to compound sulfamethoxazole, fluconazole, statins, sodium valproate, phenytoin sodium, propylthiouracil, and various traditional Chinese medicine preparations. The more hepatotoxic drugs used concurrently, the higher the risk of drug-induced liver injury. This invention incorporates the cumulative effect of combined medication use into the risk score calculation.

[0035] The sixth category is nutritional status data, specifically including serum albumin and hemoglobin levels. The normal reference range for serum albumin is 35 g / L to 50 g / L; levels below 35 g / L suggest malnutrition or impaired liver synthesis function. The normal reference range for hemoglobin varies by sex: 130 g / L to 175 g / L for men and 115 g / L to 150 g / L for women. In one embodiment of this invention, a non-linear relationship was found between hemoglobin levels and the risk of liver injury; patients with hemoglobin levels greater than or equal to 134 g / L actually had a higher risk of liver injury, which may be related to a higher red blood cell count leading to an increased metabolic burden on drugs.

[0036] Preferably, the baseline data acquisition module also collects data on the patient's previous anti-tuberculosis treatment history. Relapsed patients who have previously received anti-tuberculosis treatment have a risk of liver injury more than three times that of newly diagnosed patients, which may be related to liver sensitization to anti-tuberculosis drugs.

[0037] In one embodiment of the present invention, after baseline data acquisition is completed, the system automatically performs data integrity verification and data quality control. For missing data, different processing strategies are adopted according to the data type. Numerical continuous variables such as age, body mass index, and liver function indicators are filled with the median of the same patient group. Categorical variables such as hepatitis B surface antigen test results are filled with the most common category or the most conservative category. For example, the missing hepatitis B surface antigen status is set to positive by default to avoid missing high-risk patients.

[0038] Step S2: Baseline risk score calculation and risk stratification.

[0039] In one embodiment of the present invention, the risk prediction module receives baseline data output by the baseline data acquisition module, first performs standardized preprocessing on the baseline data, then inputs it into the liver injury risk prediction model to calculate the baseline risk score, and finally outputs risk stratification results and monitoring frequency recommendations based on the baseline risk score.

[0040] The specific implementation of standardized preprocessing is as follows. For continuous numerical variables, the min-max normalization method is used to map the original values ​​to the interval between zero and one. Let the original variable value be... The minimum value of this variable in the training dataset is The maximum value is The standardized value The calculation formula is:

[0041] ,

[0042] in: The standardized variable values ​​range from [0, 1] and are dimensionless. These are the original variable values, and the units are related to the variable type. For example, age is in years, and liver function indicators are in U / L or μmol / L. This is the minimum value of the variable in the training dataset; This represents the maximum value of the variable in the training dataset. The technical effect of this standardization method is to eliminate the differences in scale between different variables, making the contributions of each variable to the risk score comparable.

[0043] For categorical variables, one-hot encoding is used to convert them into numerical vectors. For example, the results of hepatitis B surface antigen testing include two categories: negative and positive. After conversion, a binary vector of length two is formed, with negative encoded as vector (1, 0) and positive encoded as vector (0, 1).

[0044] The liver injury risk prediction model calculates a baseline risk score based on a multi-factor weighted scoring algorithm. Let the standardized risk factor values ​​be... The corresponding weighting coefficient is The baseline risk score The calculation formula is:

[0045] ,

[0046] in: The baseline risk score ranges from [0, 100], is dimensionless, and a higher value indicates a higher risk of drug-induced liver injury. For the first The weight coefficients of each risk factor range from [0, 1], and the sum of all weight coefficients is 1. The weight coefficients are obtained by Cox regression analysis on large-scale clinical cohort data. For the first The standardized values ​​of the risk factors range from [0, 1] and are dimensionless. This represents the total number of risk factors included in the model.

[0047] In one embodiment of the present invention, based on existing clinical research evidence, the weighting coefficients of each major risk factor are set as follows: Age factor weighting The baseline alanine aminotransferase factor weight is set to 0.15. The baseline aspartate aminotransferase factor weight is set to 0.12. The weight of the hepatitis B surface antigen positive factor is set to 0.10. The weight of the drinking history factor is set to 0.18. Set to 0.12, weight of body mass index factor Set to 0.08, serum albumin factor weight Set to 0.10, and combine the weights of hepatotoxic drug factors. The weight of the previous anti-tuberculosis treatment history was set to 0.08. The weighting coefficient is set to 0.07. The above weighting coefficients are selected based on the fact that the risk of liver damage in patients with positive hepatitis B surface antigen is more than 1.5 times that of patients with negative hepatitis B surface antigen, and therefore they are given the highest weight; age and baseline liver enzyme levels are independent risk factors validated by large-scale cohort studies, and are given relatively high weights; other factors are given corresponding weights according to the magnitude of their relevant risks.

[0048] The rules for risk stratification based on baseline risk scores are as follows. This invention sets a high-risk threshold. A score of 70 points corresponds to a medium-risk threshold. It is 40 points. When, output the high-risk stratification results; when When, output the results of medium-risk stratification; when At that time, output the low-risk stratification results.

[0049] The thresholds were selected based on the following criteria. The high-risk threshold of 70 points was set based on clinical cohort validation. This threshold corresponds to a liver injury incidence rate of approximately 15% to 20%, more than three times the incidence rate in the general population, thus possessing high clinical early warning value. The medium-risk threshold of 40 points resulted in a liver injury incidence rate of approximately 8% to 12% in the intermediate-risk population, slightly higher than the average level in the general population. This requires moderately enhanced monitoring but does not necessitate the highest level of attention.

[0050] The recommended monitoring frequency is automatically generated based on risk stratification results. For high-risk patients, a weekly liver function test is recommended, specifically a combined test every 7 days for alanine aminotransferase (ALT), aspartate aminotransferase (AST), total bilirubin, and alkaline phosphatase (ALP) to ensure timely detection of early signs of liver damage. For intermediate-risk patients, a bi-weekly liver function test is recommended, balancing monitoring intensity with resource consumption. For low-risk patients, a four-week liver function test is recommended, 28 days, to avoid over-monitoring and the resulting waste of medical resources and burden on patients.

[0051] Preferably, the recommended monitoring frequency also takes into account the treatment stage. During the intensive phase of anti-tuberculosis treatment (the first two months), the risk of hepatotoxicity is highest due to the simultaneous use of four first-line drugs. Therefore, the monitoring frequency should be appropriately increased for all risk stratified patients: high-risk patients can be tested twice a week, intermediate-risk patients once a week, and low-risk patients once every two weeks. After entering the consolidation phase (from the third month onwards), the regimen is reduced to a two- or three-drug regimen with pyrazinamide, and the monitoring frequency can be appropriately reduced.

[0052] Step S3: Dynamic monitoring and Bayesian risk update.

[0053] In one embodiment of the present invention, the dynamic monitoring module periodically collects dynamic monitoring data of liver function indicators according to the monitoring frequency recommendation generated in step S2 during anti-tuberculosis treatment. The dynamic monitoring data includes the detection values ​​of alanine aminotransferase, aspartate aminotransferase, total bilirubin, and alkaline phosphatase at each monitoring time point, forming a time-series data sequence of each indicator.

[0054] Let the first The measured values ​​of each indicator during this monitoring were as follows: , , and The baseline values ​​are respectively , , and The dynamic monitoring module records the timestamp of each test and calculates the cumulative time from the start of treatment to the current test. The unit is days (d).

[0055] The trend analysis module performs in-depth analysis of dynamic monitoring data, calculating the rate of change and trend of change for each indicator. Taking alanine aminotransferase as an example, its rate of change... The calculation formula is:

[0056] ,

[0057] in: For alanine aminotransferase from baseline to the 1st The average rate of change for each monitoring session is expressed in U / (L·d). Positive values ​​indicate an upward trend, negative values ​​indicate a downward trend, and larger absolute values ​​indicate faster changes. For the first The alanine aminotransferase levels were measured during the monitoring, in U / L. The baseline alanine aminotransferase (ALT) value is given in U / L. From the start of treatment to the first The cumulative time of each monitoring session is expressed in days (d). The calculation methods for the rates of change of aspartate aminotransferase, total bilirubin, and alkaline phosphatase are the same.

[0058] The threshold values ​​for the rate of change are set as follows. In one embodiment of the present invention, the warning rate threshold for alanine aminotransferase is set to 2 U / (L·d), meaning an average daily increase of 2 U / L. This threshold implies that alanine aminotransferase will increase by approximately 28 U / L within a standard two-week monitoring interval, which is close to the upper limit of normal for patients with normal baseline levels, thus providing early warning value. The warning rate threshold for aspartate aminotransferase is also set to 2 U / (L·d). The warning rate threshold for total bilirubin is set to 1 μmol / (L·d). The warning rate threshold for alkaline phosphatase is set to 5 U / (L·d).

[0059] The trend is determined by performing linear regression analysis on multiple consecutive measurements. Let the most recent... The time point for this test is The corresponding alanine aminotransferase detection value is The least squares method was used to fit the linear regression equation. ,in is the regression slope. When When it is determined to be an upward trend, When it is determined to be a downward trend, The trend is determined to be stable. In one embodiment of the present invention, the trend determination threshold is... Set to 0.5U / (L·d).

[0060] Preferably, this invention employs a weighted linear regression method, assigning higher weights to detection values ​​at more recent time points to enhance the sensitivity of trend judgment to recent changes. The weights are calculated using an exponential decay function, assuming the time elapsed since the current detection is 1. The weight of the detection points for each day is ,in The attenuation coefficient is set to 0.05 in this invention. This setting makes the weight of the detection point 14 days ago about half that of the current detection point.

[0061] Bayesian dynamic risk update is one of the core innovative mechanisms of this invention, achieving effective integration of baseline prior risk and dynamic monitoring evidence. The implementation of the Bayesian update algorithm is as follows.

[0062] Baseline risk score Convert to prior risk probability :

[0063] ,

[0064] in: The prior risk probability, with a value range of [0, 1], represents the probability of liver injury predicted based solely on baseline features.

[0065] Calculate the strength of risk evidence based on dynamic monitoring data. Taking into account the relative deviation and rate of change of each liver function indicator:

[0066] ,

[0067] in: For the first The strength of risk evidence at each monitoring point is dimensionless; positive values ​​indicate evidence of increased risk, and negative values ​​indicate evidence of decreased risk. For the first The weights of the liver function indicators are as follows: alanine aminotransferase weight is 0.35, aspartate aminotransferase weight is 0.25, total bilirubin weight is 0.25, and alkaline phosphatase weight is 0.15. For the first The first indicator in the The measured value at the time of the next monitoring; For the first The baseline value of each indicator; For the first The upper limit of the normal reference value for each indicator; For the first The rate of change of each indicator; For the first The warning rate threshold for each indicator; The relative contribution coefficient of the rate of change is set to 0.3 in this invention. This parameter setting ensures that the rate of change contributes approximately 30% of the total contribution to the risk evidence.

[0068] Convert the strength of risk evidence into likelihood ratio :

[0069] ,

[0070] in: The likelihood ratio ranges from (0, +∞). A value greater than 1 indicates that the observed monitoring data supports the hypothesis of increased risk, while a value less than 1 indicates that it supports the hypothesis of decreased risk. The Bayesian update coefficient controls the influence of monitoring data on risk assessment, with a value range of [0.1, 0.5]. This invention dynamically adjusts the coefficient based on the time elapsed between the monitoring time point and the start of treatment. value.

[0071] Bayesian update coefficients The dynamic adjustment rules are as follows. In the initial stage of treatment (first 2 weeks)... Setting it to 0.2 is more conservative in responding to single monitoring data, avoiding fluctuations in risk assessment due to initial adaptive changes; this is particularly relevant during the middle of treatment (weeks 3 to 8). Gradually increase to 0.4, making full use of accumulated monitoring data to update the risk assessment; in the later stages of treatment (starting from week 9). Stabilizing at 0.3 balances responsiveness to new data with the stability of the assessment.

[0072] Calculating posterior risk probability using Bayes' theorem :

[0073] ,

[0074] in: The posterior risk probability, with a value range of [0, 1], represents the predicted probability of liver injury after combining baseline features and dynamic monitoring data.

[0075] Transform posterior risk probability into dynamic risk score :

[0076] ,

[0077] in: The dynamic risk score has a value range of [0, 100], is dimensionless, and has the same meaning as the baseline risk score, but incorporates dynamic monitoring information.

[0078] The technical advantages of the Bayesian dynamic risk update mechanism are as follows: First, it achieves the organic integration of baseline risk priors and dynamic monitoring evidence, enabling risk assessment to consider both the patient's inherent characteristics and timely reflect changes during the treatment process; Second, it uses a likelihood ratio framework to provide a clear statistical interpretation of risk updates, making it easier for clinicians to understand and trust; Third, by dynamically adjusting the update coefficient, it enables the system to have appropriate response characteristics at different stages of treatment, avoiding both oversensitivity and response lag.

[0079] Preferably, when the dynamic risk score changes significantly compared to the baseline risk score, the system automatically adjusts the recommended monitoring frequency. The specific rule is: if... If the dynamic risk score increases by more than 20 points compared to the baseline risk score, the monitoring frequency will be increased by one level, for example, from once every four weeks to once every two weeks; if If the dynamic risk score decreases by more than 15 points compared to the baseline risk score, the monitoring frequency can be reduced by one level, but the monitoring frequency for high-risk patients should not be less than once every two weeks. This adaptive adjustment mechanism forms a closed-loop control, realizing the dynamic optimization of the monitoring strategy.

[0080] Step S4: Early warning judgment and graded alarm.

[0081] In one embodiment of the present invention, the early warning judgment module comprehensively judges the early signs of liver damage based on the degree and rate of change of liver function indicators from the baseline, and outputs a graded early warning alert. The early warning judgment adopts a multi-indicator collaborative analysis mechanism, overcoming the limitations of single-indicator judgment.

[0082] Multi-indicator co-variance The calculation method is as follows. First, calculate the relative deviation of each liver function indicator:

[0083] ,

[0084] ,

[0085] ,

[0086] ,

[0087] in: , , , These are the relative deviations of alanine aminotransferase, aspartate aminotransferase, total bilirubin, and alkaline phosphatase, respectively. They are dimensionless and represent the normalized result of the deviation of the current value from the baseline value, using the upper limit of normal as a scale. , , , The first The measured values ​​of each indicator during the next monitoring session; , , , These are the baseline values ​​for each indicator; , , , These are the upper limits of the normal reference values ​​for each indicator, with values ​​of 40 U / L, 40 U / L, 21 μmol / L, and 130 U / L, respectively.

[0088] Then, the relative deviations of each indicator are weighted and summed to obtain the multi-indicator coordinated deviation:

[0089] ,

[0090] in: This is a dimensionless, multi-indicator deviation measure; a larger value indicates a more severe overall deviation of liver function indicators from the normal state. The weighting is based on the following: alanine aminotransferase is the most sensitive indicator of hepatocellular damage and is assigned the highest weight; aspartate aminotransferase and total bilirubin are of significant value in assessing the severity of liver damage and are assigned a medium weight; alkaline phosphatase mainly reflects cholestasis and has relatively low specificity in anti-tuberculosis drug-induced liver injury, so it is assigned a lower weight.

[0091] The tiered warning system employs multi-condition logic for judgment. This invention sets four warning levels: red, orange, yellow, and no warning, corresponding to different risk states from severe to normal.

[0092] A red alert is triggered if any of the following conditions are met: First, the alanine aminotransferase (ALT) level is greater than or equal to five times the upper limit of the normal reference range. ,correspond Second, alanine aminotransferase levels greater than or equal to three times the upper limit of the normal reference range, coupled with total bilirubin levels greater than or equal to twice the upper limit of the normal reference range, i.e. and ,correspond and A red alert indicates that drug-induced liver injury has occurred and immediate intervention is required.

[0093] The triggering conditions for an orange alert are: an alanine aminotransferase (ALT) level greater than or equal to three times but less than five times the upper limit of the normal reference range, without a significant increase in total bilirubin. and ,correspond and An orange alert indicates that liver enzymes are significantly elevated but have not yet reached the criteria for severe liver damage, requiring close monitoring and consideration of adjusting the treatment plan.

[0094] A yellow alert is triggered if any of the following conditions are met: First, the alanine aminotransferase (ALT) level is greater than or equal to 1.5 times but less than 3 times the upper limit of the normal reference range. ,correspond Second, if the rate of change of any liver function indicator exceeds a preset rate threshold, that is... or or or Third, the deviation of multiple indicators from a predetermined threshold is greater than the threshold value. A yellow alert indicates early signs of liver damage, requiring increased monitoring frequency and the initiation of preventative interventions.

[0095] If none of the above warning conditions are met, a no-warning status will be output, indicating that liver function indicators are within an acceptable range and the current monitoring and treatment strategy can be maintained.

[0096] Preferably, the early warning judgment module also calculates an R value based on the relative increase fold of alanine aminotransferase and alkaline phosphatase to determine the type of liver injury. The formula for calculating the R value is:

[0097] ,

[0098] in: R value for determining the type of liver injury, dimensionless; This is the current measured value of alanine aminotransferase, in U / L; The upper limit of the normal reference value for alanine aminotransferase is 40 U / L. This is the current measured value of alkaline phosphatase, in U / L; The upper limit of the normal reference value for alkaline phosphatase is 130 U / L.

[0099] The rule for determining the type of liver injury is: when When diagnosed as hepatocellular injury, it indicates that hepatocellular necrosis is the primary feature; when When diagnosed as cholestatic injury, it indicates that the primary cause is bile excretion obstruction; when The injury was initially classified as mixed-type, indicating the simultaneous presence of hepatocellular damage and cholestasis. Different types of liver injury have different clinical prognoses and treatment principles. Hepatocellular damage is usually associated with isoniazid and pyrazinamide, while cholestatic damage is usually associated with rifampin. The classification result provides a basis for adjusting the subsequent treatment plan.

[0100] Step S5: Assessment report generation and treatment recommendations.

[0101] In one embodiment of the present invention, the assessment report generation module generates a structured liver injury risk assessment report based on the risk stratification results, dynamic risk scores, trend analysis results, and early warning judgment results, and generates individualized treatment plan adjustment suggestions based on the early warning level and liver function index change characteristics.

[0102] The liver injury risk assessment report includes the following core modules: Module 1 is a summary of basic patient information, including patient identification, age, gender, type of tuberculosis, treatment plan, and current treatment stage. Module 2 is a risk assessment summary, including baseline risk score, current dynamic risk score, risk stratification results, and risk trend. Module 3 is a summary of liver function indicator monitoring, including baseline values, latest measured values, magnitude of change, rate of change, and trend graphs for each indicator. Module 4 is a description of the warning status, including the current warning level, the specific conditions triggering the warning, and the liver injury type assessment results. Module 5 is treatment recommendations, including specific management measures suggested for the current condition.

[0103] Treatment plan adjustment suggestions are generated using a tiered response strategy based on the warning alert level. When a red alert is issued, the system generates the following suggestions: First, it is recommended to immediately discontinue all anti-tuberculosis drugs; second, it is recommended to initiate liver-protective therapy, with preferred drugs including reduced glutathione, polyene phosphatidylcholine, and magnesium isoglycyrrhizinate; third, it is recommended to recheck liver function indicators every 2 to 3 days until significant improvement is achieved; fourth, it is recommended to consult with a hepatologist or infectious disease specialist to assess the condition; fifth, after liver function recovers, it is recommended to restart anti-tuberculosis therapy according to the principle of introducing drugs one by one, prioritizing ethambutol, which has lower hepatotoxicity, followed by rifampin or rifapentine, isoniazid, and finally considering whether to reintroduce pyrazinamide.

[0104] When an orange alert is issued, the system generates the following recommendations: First, it is recommended to increase the monitoring frequency to once every 3 days; second, it is recommended to add hepatoprotective drugs for preventive intervention; third, based on the type of liver injury, if it is hepatocellular injury, it is recommended to prioritize discontinuing pyrazinamide and isoniazid, and if it is cholestatic injury, it is recommended to prioritize discontinuing rifampin; fourth, if liver function indicators continue to rise within 1 week after enhanced monitoring and hepatoprotective treatment, it is recommended to upgrade to a red alert response.

[0105] When a yellow alert is issued, the system generates the following recommendations: First, it is recommended to increase the monitoring frequency by one level, for example, from once every four weeks to once every two weeks; Second, it is recommended to initiate preventive liver protection measures, such as oral hepatoprotective drugs like silymarin or bicyclol; Third, it is recommended that patients strictly abstain from alcohol and avoid using other hepatotoxic drugs; Fourth, it is recommended to strengthen patient education, informing them of the early symptoms of liver damage such as fatigue, loss of appetite, nausea, and dark urine, and to seek medical attention promptly if symptoms appear.

[0106] When no warning is issued, the system generates the following recommendations: First, it is recommended to maintain the current anti-tuberculosis treatment regimen; second, it is recommended to continue to regularly monitor liver function according to the monitoring frequency corresponding to the risk stratification; third, it is recommended that the patient maintain a healthy lifestyle and avoid drinking alcohol and using unnecessary hepatotoxic drugs.

[0107] Preferably, the assessment report generation module also generates a risk trend analysis chart and prognostic prediction based on the patient's cumulative monitoring data. The risk trend analysis chart uses time as the horizontal axis and dynamic risk score and various liver function indicators as the vertical axis to visually display the evolution trajectory of risk status during treatment. The prognostic prediction extrapolates the expected risk status within a certain future time window based on the current risk trend, providing a prospective reference for clinical decision-making.

[0108] This invention also provides a system for predicting and monitoring the risk of liver injury during tuberculosis anti-tuberculosis treatment, such as... Figure 2 As shown, the system includes a baseline data acquisition module 1, a risk prediction module 2, a dynamic monitoring module 3, an early warning judgment module 4, and an assessment report generation module 5. The modules communicate and collaborate with each other through data interfaces.

[0109] The baseline data acquisition module 1 is used to collect patients' baseline characteristic data before the initiation of anti-tuberculosis treatment. This module interfaces with the hospital information system, laboratory information system, and electronic medical record system to automatically acquire structured data such as patient demographics, basic liver function indicators, and viral infection status. For lifestyle data such as alcohol and smoking history, as well as concomitant medication data, the baseline data acquisition module 1 provides a standardized electronic questionnaire interface for medical staff or patients to input data. The baseline data acquisition module 1 also has a data integrity verification function, prompting for missing data and providing reasonable default value filling options. The baseline data acquisition module outputs a standardized format baseline data package as input to the risk prediction module.

[0110] The risk prediction module 2 calculates a baseline risk score based on baseline data and outputs risk stratification results. It incorporates a liver injury risk prediction model implemented using a multi-factor weighted scoring algorithm. Model parameters are obtained through statistical analysis of large-scale clinical cohort data. The risk prediction module 2 receives the baseline data package output from the baseline data acquisition module 1. It first performs data preprocessing, including standardization and encoding conversion, then calls the risk prediction model to calculate the baseline risk score. Finally, it outputs high-risk, intermediate-risk, or low-risk risk stratification results based on a comparison of the risk score with preset thresholds. The risk prediction module 2 also queries the monitoring strategy rule base based on the risk stratification results to generate corresponding monitoring frequency recommendations. The risk prediction module outputs a risk assessment result package, containing the baseline risk score, risk stratification results, and monitoring frequency recommendations, which serves as input to the dynamic monitoring module and the assessment report generation module.

[0111] The dynamic monitoring module 3 is used to collect and analyze dynamic monitoring data of liver function indicators during treatment. The dynamic monitoring module 3 interfaces with the laboratory information system and automatically acquires liver function test results according to the recommended monitoring frequency settings. The dynamic monitoring module 3 has a built-in trend analysis module, which uses the rate of change calculation formula and trend judgment algorithm described in the method embodiments to analyze the time-series data of each liver function indicator. The dynamic monitoring module also has a built-in Bayesian risk update engine. This sub-module receives the baseline risk score output by the risk prediction module as a prior probability, combines it with the risk evidence strength calculated from the dynamic monitoring data, and outputs a dynamic risk score using a Bayesian update algorithm. When the dynamic risk score changes significantly compared to the baseline risk score, the dynamic monitoring module automatically adjusts the recommended monitoring frequency and notifies relevant clinical personnel. The dynamic monitoring module 3 outputs a dynamic monitoring analysis result package, including the rate of change, trend, dynamic risk score, and updated monitoring frequency recommendation for each indicator.

[0112] The early warning judgment module 4 is used to comprehensively judge early signs of liver injury and output graded early warning alerts. The early warning judgment module 4 receives the dynamic monitoring analysis result package output by the dynamic monitoring module 3, and uses the multi-indicator synergistic deviation calculation formula and graded early warning judgment rules described in the method embodiment to assess the current liver function status and determine the warning level. The early warning judgment module 4 also calculates the R value to determine the type of liver injury. When a yellow or higher level warning is triggered, the early warning judgment module sends a real-time warning notification to the attending physician via system message, SMS, or mobile application push. The early warning judgment module 4 outputs an early warning judgment result package, including the warning level, triggering condition description, multi-indicator synergistic deviation, and liver injury type judgment result.

[0113] The assessment report generation module 5 generates liver injury risk assessment reports and treatment plan adjustment recommendations. This module receives the output packages from the risk prediction module 2, dynamic monitoring module 3, and early warning judgment module 4, integrates the information according to a predefined report template, and generates a structured assessment report document. The module also includes a built-in treatment recommendation rule engine. This sub-engine queries the treatment recommendation knowledge base based on the early warning level and liver injury type to generate individualized treatment plan adjustment recommendations. Furthermore, the assessment report generation module 5 has a report export function, supporting the output of assessment reports in electronic document format for clinical archiving and patient access.

[0114] In one embodiment of the present invention, the system is deployed on the information platform of a medical institution and integrated with existing hospital information systems, laboratory information systems, and electronic medical record systems. The system adopts a browser / server architecture, with a web-based user interface provided at the front end for clinicians and nurses to operate, and a risk prediction model and data processing engine deployed at the back end. The system supports concurrent access by multiple users and has user access management and operation log auditing functions to ensure data security and system traceability.

[0115] In another embodiment of the invention, the system is deployed as a mobile application, supporting patients to independently input symptom information and view assessment reports, enabling collaborative liver injury monitoring and management between doctors and patients. The mobile application interfaces with the medical institution's system through a secure interface to ensure data synchronization and privacy protection.

[0116] To verify the technical effectiveness of the method and system of this invention, a retrospective cohort validation study was conducted at a tertiary-level Class A infectious disease hospital. The study included 1200 patients with pulmonary tuberculosis who received anti-tuberculosis treatment between January 2020 and June 2024, among whom 127 cases of drug-induced liver injury occurred, with an incidence rate of 10.6%.

[0117] The method of this invention was used to perform risk prediction and monitoring simulations for all patients. Regarding baseline risk scores, the average baseline risk score for patients with liver injury was 62.3 points, while the average baseline risk score for patients without liver injury was 38.7 points; the difference between the two groups was statistically significant. In terms of risk stratification, the incidence of liver injury was 22.4% in the high-risk group, 9.8% in the intermediate-risk group, and 3.2% in the low-risk group, showing a significant stratification effect.

[0118] Validation of the Bayesian dynamic risk update showed that in patients with liver injury, the dynamic risk score began to rise significantly on average 8.5 days before the onset of liver injury, providing an earlier warning opportunity than methods relying solely on static baseline risk scores. The sensitivity of the warning assessment was 91.3%, the specificity was 85.6%, the positive predictive value was 42.8%, and the negative predictive value was 98.9%.

[0119] Regarding the optimization of monitoring resources, the adaptive monitoring strategy of this invention reduces the average number of monitoring sessions for low-risk patients from 8 to 4, a 50% reduction in monitoring resource consumption, while ensuring no cases of liver injury are missed. Monitoring density for high-risk patients is increased, and the early liver injury identification rate is improved by 35%.

[0120] In summary, the method and system for predicting and monitoring the risk of liver injury during tuberculosis treatment provided by this invention, through innovative mechanisms such as deep fusion of baseline data and dynamic monitoring data, Bayesian dynamic risk updating, and multi-indicator synergistic early warning judgment, achieves accurate prediction, early identification, and individualized monitoring and management of drug-induced liver injury risk, and has significant clinical application value.

[0121] The embodiments of the present invention are not limited to the specific embodiments described above. Those skilled in the art can make various equivalent changes or substitutions based on the technical solutions of the present invention, and all such changes or substitutions should be included within the protection scope of the present invention.

Claims

1. A method for predicting and monitoring the risk of liver injury during anti-tuberculosis treatment in tuberculosis patients, characterized in that, Includes the following steps: Step S1: Collect baseline data of tuberculosis patients. The baseline data includes demographic data, basic liver function indicators, viral infection status data, lifestyle data, concomitant medication data, and nutritional status data. Specifically, the demographic data includes age, sex, and body mass index; the basic liver function indicators include baseline values ​​of alanine aminotransferase (ALT), aspartate aminotransferase (AST), total bilirubin, and alkaline phosphatase; the viral infection status data includes hepatitis B surface antigen (HBsAg) and hepatitis C antibody (HCA) test results; the lifestyle data includes alcohol consumption history and smoking history; the concomitant medication data includes the types and quantities of hepatotoxic drugs used concurrently; and the nutritional status data includes serum albumin and hemoglobin levels. Step S2: Standardize and preprocess the baseline data, input the preprocessed baseline data into the liver injury risk prediction model, the liver injury risk prediction model calculates the baseline risk score based on the multi-factor weighted scoring algorithm, and outputs the three-level risk stratification results of high risk, medium risk and low risk based on the comparison of the baseline risk score with the preset risk threshold, and generates corresponding monitoring frequency suggestions based on the risk stratification results. Step S3: During anti-tuberculosis treatment, dynamic monitoring data of liver function indicators are collected regularly according to the recommended monitoring frequency. The dynamic monitoring data includes time-series detection values ​​of alanine aminotransferase, aspartate aminotransferase, total bilirubin, and alkaline phosphatase. The rate of change and trend of each indicator in the dynamic monitoring data relative to the baseline value are calculated using the trend analysis module. Based on the dynamic monitoring data, the baseline risk score is dynamically updated using a Bayesian update algorithm to obtain the dynamic risk score. Step S4: The early warning judgment module calculates the multi-indicator synergistic deviation degree based on the degree of deviation of each liver function indicator from the baseline, and combines the change rate of each indicator to comprehensively judge the early signs of liver damage. Based on the judgment result, it outputs graded early warnings including red warning, orange warning, yellow warning and no warning.

2. The method according to claim 1, characterized in that, In step S2, the rule for generating the recommended monitoring frequency is as follows: when the risk stratification result is high-risk, the monitoring frequency is once a week; when the risk stratification result is medium-risk, the monitoring frequency is once every two weeks; and when the risk stratification result is low-risk, the monitoring frequency is once every four weeks.

3. The method according to claim 1, characterized in that, In step S2, the risk threshold includes a high-risk threshold and a medium-risk threshold. When the baseline risk score is greater than or equal to the high-risk threshold, the high-risk stratification result is output. When the baseline risk score is less than the high-risk threshold and greater than or equal to the medium-risk threshold, the medium-risk stratification result is output. When the baseline risk score is less than the medium risk threshold, the low-risk stratification result is output.

4. The method according to claim 1, characterized in that, In step S3, the rate of change is calculated as follows: the difference between the current detection value and the baseline value is calculated and divided by the time interval from the start of treatment to the current detection to obtain the average rate of change of the index. The trend of change is determined by linear regression analysis of the detected values. When the regression slope is positive, it is determined to be an upward trend; when the regression slope is negative, it is determined to be a downward trend; and when the absolute value of the regression slope is less than a preset trend threshold, it is determined to be a stable trend.

5. The method according to claim 1, characterized in that, In step S3, the Bayesian update algorithm is implemented as follows: the baseline risk score is used as the prior risk probability, the risk evidence corresponding to the dynamic monitoring data is used as the input of the likelihood function, the posterior risk probability is calculated through Bayes' theorem, and the posterior risk probability is converted into a dynamic risk score; wherein, the Bayesian update coefficient is dynamically adjusted according to the time length between the monitoring time point and the start of treatment.

6. The method according to claim 1, characterized in that, In step S4, the calculation method for the multi-indicator synergistic deviation is as follows: the relative deviations of alanine aminotransferase, aspartate aminotransferase, total bilirubin and alkaline phosphatase are calculated respectively, and the weighted sum of the relative deviations of each indicator is used to obtain the multi-indicator synergistic deviation; wherein, the relative deviation of each indicator is the difference between the current detection value and the baseline value divided by the upper limit of the normal reference value.

7. The method according to claim 1, characterized in that, In step S4, the judgment rules for the graded warning are as follows: when alanine aminotransferase (ALT) rises to more than five times the upper limit of the normal reference value, or when ALT rises to more than three times the upper limit of the normal reference value and is accompanied by total bilirubin rising to more than twice the upper limit of the normal reference value, a red warning is issued; when ALT rises to between three and five times the upper limit of the normal reference value, an orange warning is issued; when ALT rises to between 1.5 and 3 times the upper limit of the normal reference value, or when the rate of change of any liver function indicator exceeds a preset rate threshold, a yellow warning is issued; otherwise, no warning is issued.

8. The method according to claim 1, characterized in that, Step S4 further includes: calculating the R value based on the relative increase of alanine aminotransferase and alkaline phosphatase, and determining the type of liver injury based on the R value. When the R value is greater than or equal to five, it is determined to be hepatocellular injury; when the R value is less than or equal to two, it is determined to be cholestatic injury; and when the R value is greater than two and less than five, it is determined to be mixed injury.

9. The method according to claim 1, characterized in that, The process also includes step S5, where the assessment report generation module generates a liver injury risk assessment report based on the risk stratification results, dynamic risk score, trend analysis results, and early warning judgment results. It also generates treatment plan adjustment suggestions based on the early warning level and liver function indicator changes. These suggestions include: when a red warning is issued, it is recommended to immediately discontinue anti-tuberculosis drugs and initiate liver-protective treatment; when an orange warning is issued, it is recommended to closely monitor the patient and consider reducing the dosage or switching to a drug regimen with lower hepatotoxicity; when a yellow warning is issued, it is recommended to increase the monitoring frequency and initiate preventative liver-protective measures; and when no warning is issued, it is recommended to maintain the current treatment plan and monitoring frequency.

10. A system for predicting and monitoring the risk of liver injury during tuberculosis anti-tuberculosis treatment, used to implement the method of claim 9, characterized in that, include: The baseline data acquisition module is used to collect baseline data of tuberculosis patients. The baseline data includes demographic data, basic liver function index data, viral infection status data, lifestyle data, combined medication data, and nutritional status data. The risk prediction module is used to standardize and preprocess the baseline data, input the preprocessed baseline data into the liver injury risk prediction model to calculate the baseline risk score, output the three-level risk stratification results of high risk, intermediate risk and low risk based on the baseline risk score, and generate corresponding monitoring frequency suggestions based on the risk stratification results. The dynamic monitoring module is used to collect dynamic monitoring data of liver function indicators according to the recommended monitoring frequency during anti-tuberculosis treatment. The module calculates the trend and rate of change of each indicator through the trend analysis module, and dynamically updates the baseline risk score based on the dynamic monitoring data using a Bayesian update algorithm. The early warning judgment module is used to calculate the synergistic deviation of multiple indicators based on the degree of deviation of liver function indicators from the baseline, and to comprehensively judge the early signs of liver damage by combining the rate of change of each indicator, and output graded early warnings. The assessment report generation module is used to generate liver injury risk assessment reports and treatment plan adjustment suggestions based on risk stratification results, dynamic risk scores, and early warning judgment results.

Images