Risk prediction system and method for lung cancer treatment immune adverse events

Abstract

The invention provides a risk prediction system and method for lung cancer treatment immune adverse events, and belongs to the field of artificial intelligence assisted medical decision. The system comprises an information acquisition module used for acquiring patient information from a hospital information system and a patient self-reporting platform; the risk assessment module is used for predicting the occurrence risk level of the immune-related adverse events through a machine learning model based on the collected information; the report generation module is used for generating three types of personalized reports for doctors, clinical pharmacists and patients according to evaluation results; the report sending module is used for sending the report to each terminal; and the central management module is used for data verification, consistency check and real-time synchronization. According to the invention, early and accurate prediction and multi-role collaborative management of adverse risks in the immunotherapy process of the lung cancer patient are realized, and the treatment safety and medical collaborative efficiency are improved.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence-assisted medical decision-making technology, and in particular relates to a risk prediction system and method for adverse immune events in lung cancer treatment. Background Technology

[0002] In recent years, tumor immunotherapy targeting immune checkpoints has attracted much attention due to its significant efficacy. Programmed death-1 (PD-1) inhibitors, as a representative drug, have demonstrated excellent efficacy in lung cancer treatment and have been recommended as first- or second-line treatments in the guidelines of the Chinese Society of Clinical Oncology (CSCO) and the National Comprehensive Cancer Network (NCCN). With the continuous development of PD-1 inhibitors and the improvement of drug accessibility, reports of their adverse reactions are also increasing. This off-target inflammatory response is clinically classified as "immune-related adverse events" (irAEs), which may lead to treatment interruption or even death. The overall incidence of irAEs in clinical trials ranges from 27% to 78%, with cardiotoxicity and interstitial pneumonia among irAEs having a mortality rate as high as 45%. Therefore, early identification of potential irAE populations is crucial for improving patient survival benefits. However, the clinical characteristics of irAEs are often subtle, making early diagnosis difficult. Therefore, developing prevention and management of irAEs has become a key strategy for optimizing patient treatment outcomes and reducing toxicity risks. Summary of the Invention

[0003] The purpose of this invention is to solve the problems of low accuracy and difficulty in predicting immune-related adverse events in the prior art.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: A risk prediction system for immune-related adverse events in lung cancer treatment includes: The information collection module is used to collect patient information from the hospital information system and the patient self-reporting platform; The risk assessment module is used to predict the risk level of patients developing irreversible adverse events (irAEs) based on the collected information and through machine learning models. The report generation module is used to generate three types of reports based on the risk assessment results, for physicians, clinical pharmacists, and patients. The report sending module is used to send three types of reports to the doctor's terminal, the clinical pharmacist's terminal, and the patient's terminal, respectively. The central management module is used to perform data verification, consistency checks, and real-time synchronization of the collected information and generated reports.

[0005] A method for predicting the risk of immune-related adverse events in lung cancer treatment includes the following steps: S1: Collect patient information, including demographic data, tumor characteristics, and laboratory indicators; S2: Establish a nomogram prediction model based on multivariate logistic regression analysis to quantify the weight of each risk factor; S3: Perform consistency checks on the collected patient information and update patient medication data in real time; S4: Use the RF model to predict the probability of irAEs occurring, generate and send three types of reports; S5: When the data in any of the three types of reports changes, the report is updated synchronously and cross-validated.

[0006] Furthermore, the demographic data, tumor characteristics, and laboratory indicators in step S1 are as follows: Demographic characteristics: age, sex, BMI, ECOG score, underlying diseases; Tumor-related indicators: pathological type, TNM stage, treatment method; Laboratory test indicators: platelet count, neutrophil count, lymphocyte count, eosinophil count, monocyte count, serum albumin, creatinine, liver function indicators, lactate dehydrogenase, creatine kinase; Derived nutritional inflammation indices: NLR, PLR, MLR, PNI, SII.

[0007] Furthermore, the weight calculation formula in step S2 is as follows:

[0008] in, They are professions ,job title Relevance to major The weighting coefficients satisfy the condition. .

[0009] Furthermore, the consistency check in step S3 includes checking required fields, data types and ranges, patient ID, treatment stage, and follow-up time.

[0010] Furthermore, the three types of reports in step S4 are as follows: First report: Sent to the doctor's terminal, including the risk level of irAEs, matching adverse event warning signals, graded treatment procedures and emergency response strategies; The second report is sent to the clinical pharmacist's terminal and includes the pharmaceutical care plan, medication education content, follow-up schedule, and abnormal handling procedures. The third report is sent to the patient's terminal and includes self-monitoring guidelines, lifestyle suggestions, follow-up arrangements, and symptom recording methods.

[0011] The beneficial effects of this invention are as follows: By utilizing an immune adverse event risk prediction system, dynamic monitoring is conducted at each cycle of PD-1 inhibitor therapy to accurately identify potential patients at an early stage of irAEs. This assists physicians in taking timely preventative measures and flexibly adjusting treatment plans, while also facilitating targeted medication monitoring and education by clinical pharmacists. Through this comprehensive intervention, the ultimate goal is to improve patient survival benefits. Attached Figure Description

[0012] Figure 1 This is the ROC curve of the training set prediction model provided by this invention; Figure 2 This is the ROC curve of the test set prediction model provided by this invention; Figure 3 This is the DCA curve of the irAEs prediction model provided by this invention; Figure 4 This is the calibration curve of the irAEs risk prediction model provided by this invention; Figure 5 This is a nodal plot of the irAEs risk prediction model provided by the present invention; Figure 6 This invention provides a risk prediction system for adverse immune events in lung cancer treatment. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0014] The application principle of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0015] Example 1: A risk prediction system for adverse immune events in lung cancer treatment includes: an information collection module, a risk assessment module, a report generation module, a report sending module, and a central management module.

[0016] The information collection module mainly obtains information by connecting with the hospital's HIS system and patient self-reporting. The risk assessment module and irAEs prediction system assess the patient's risk of irAEs during the treatment period. The report generation module generates reading reports suitable for physicians, clinical pharmacists, and patients based on the evaluation results of evidence-based assessments. The evidence-based assessments include relevant guidelines for irAEs, expert consensus, and the latest high-quality literature.

[0017] The report sending module generates a first report, a second report, and a third report, and sends the three different reports to the receiving terminals of doctors, clinical pharmacists, and patients, respectively. The central management module, which is connected to the report generation module, is used to verify the first, second, and third reports generated from the initial information. This includes verifying all data before it is entered using pre-set data verification rules and procedures, as well as performing consistency checks on patient ID, treatment stage, and follow-up time.

[0018] A method for predicting the risk of adverse immune events in lung cancer treatment, including S1. The information collection module connects to the Hospital Information System (HIS) and the mobile APP to collect patient information through patient self-reporting, including demographic characteristics such as age, gender, BMI, ECOG, and underlying diseases; tumor-related indicators such as TNM stage and treatment methods; and laboratory tests and derived nutritional and inflammatory indices such as platelet count, neutrophil count, lymphocyte count, eosinophil count, monocyte count, serum albumin, creatinine, aspartate aminotransferase, alanine aminotransferase, bilirubin, lactate dehydrogenase, creatine kinase, NLR, PLR, MLR, PNI, and SII.

[0019] S2. For samples of known adverse events related to treatment with the given formulation, multivariate analysis was performed on the pathological type of the tumor, whether an ECOG score was obtained, and the nutritional inflammation indices NLR, PNI, and SII. The weight scores of each factor were quantified, and a nomogram prediction model based on multivariate logistic regression analysis was established. Based on the independent risk factors (such as pathological type, ECOG, NLR, PNI, and SII) determined by multivariate logistic regression analysis, a random forest (RF) prediction model was constructed and trained.

[0020] The ECOG score is a widely used clinical indicator for assessing a patient's physical function and activity level. Patients with high scores often have difficulty tolerating irreversible adverse events (irAEs), and ICIs are not recommended for treatment. The score ranges from 0 to 5 points. Specifically: 0 points: Completely normal activity, able to perform all daily activities without limitation; 1 point: Mild symptoms or signs, able to perform light physical activity, including general housework or office work, but unable to engage in heavy physical labor; 2 points: Able to walk freely and take care of oneself, but daytime activity time does not exceed 50%, able to take care of oneself partially, and needs to be bedridden or sit still for more than half of the day; 3 points: Only able to take care of oneself partially, and needs to be bedridden or sit still for more than half of the day; 4 points: Bedridden, completely unable to take care of oneself; 5 points: Death.

[0021] The quantitative indicators for each factor are as follows: Occupation A is represented by a value between 0 and 1, with 1 indicating perfect occupational relevance. Professional title B is represented by a value between 0 and 1, with 1 indicating a senior professional title. Professional relevance C is represented by a value between 0 and 1, with 1 indicating perfect professional relevance.

[0022] Establish the weight calculation formula:

[0023] in, These are the weighting coefficients for occupation, professional title, and professional relevance, respectively, which must meet the conditions. To ensure a reasonable allocation of weights, the Delphi method can be used, combined with historical data analysis, to determine the importance of each factor and adjust accordingly. The value, or add new weighted quantitative indicators.

[0024] S3. Conduct consistency checks on the information input into the HIS (Hospital Information System) (obtained demographic data, tumor characteristics, laboratory tests, and derived nutritional and inflammatory indices related to patients' self-reports) to ensure the accuracy of the input patient data; and dynamically monitor and update the records of patients' medication data for each cycle in real time. The consistency check includes: checking required fields, data types, and ranges to ensure data integrity and accuracy; performing consistency checks on patient ID, treatment stage, and follow-up time; and ensuring consistency across different reports to prevent input errors.

[0025] S4. Based on the RF model, predict the potential side effect risk level of immunotherapy in patients' information, analyze the symptoms reported by patients themselves, identify and match immune-related adverse events, compare the adverse event risk with the actual risk, dynamically adjust the assessment method, and update it regularly to ensure the accuracy of the system assessment. Use the RF model to predict the probability of occurrence of immune-related adverse events and obtain the treatment assessment results; based on the treatment assessment results, generate a first report, a second report, and a third report, and send the three different reports to the receiving terminals of doctors, clinical pharmacists, and patients, respectively. S5. When the data in any of the first, second, or third reports changes, the changed data will be synchronized to other reports containing related data to ensure that the information is updated in real time. Cross-validation will be used to check the data in the first, second, or third reports, and any inconsistencies will be promptly notified to the administrator for correction.

[0026] The first report is sent to the physician's terminal and includes at least the risk level of immune-related adverse events matched in the treatment assessment results, the management rules based on evidence-based evidence, and the laboratory test results. The second report is sent to the clinical pharmacist's terminal. The second report should include at least the schedule of the pharmaceutical care plan, medication education, regular follow-up, and the corresponding handling procedures for any abnormalities found during the follow-up. The third report is sent to the patient's terminal. The third report should include at least the vital signs, the time of eating, exercising and sleeping that need to be recorded, as well as the time and place of follow-up.

[0027] This solution utilizes information technology to achieve deep collaboration among physicians, clinical pharmacists, and patients, assisting in predicting the probability of irreversible adverse events (irAEs) during future disease progression, identifying potential high-risk groups for irAEs in the early stages, and effectively ensuring the safety of immunotherapy medication for cancer patients.

[0028] For doctors, clinical pharmacists, and patients, a precise, personalized, and highly collaborative solution has been implemented in the computer system. By sending customized reports to the terminals of doctors, clinical pharmacists, and patients, communication among the three parties has been strengthened, the synergy of medical services has been improved, patients have received continuous and comprehensive management throughout the treatment process, and the efficiency and synergy of information exchange among the multiple parties have been greatly improved.

[0029] Specifically, generating the first, second, and third reports includes the following steps: When generating the first report, extract information including early symptom identification, tiered treatment procedures, and emergency response measures; when generating the second report, extract information including detailed monitoring procedures, medication education content, follow-up time points, follow-up content, and follow-up treatment for failure to follow up on time; when generating the third report, extract information including dietary, exercise, and sleep recommendations, assessment frequency, assessment tools, and recording and reporting methods. The report focuses on identifying and listing early warning signs of various adverse events (irAEs) in patients undergoing immunotherapy at high risk. These warnings are systematically matched and graded according to severity. For different irAEs, the report provides specific management information. For example, mild skin and gastrointestinal reactions (irAEs) with relatively obvious initial symptoms can be treated with symptomatic medications under professional guidance. For more severe irAEs such as immune-related myocarditis, the report details emergency response strategies, such as 24-hour monitoring, daily monitoring of key indicators, and switching immune checkpoint inhibitors, to ensure patient safety.

[0030] The second report focuses on pharmaceutical services. Before treatment, pharmacists need to comprehensively assess the patient's medication history, underlying diseases, and baseline examinations to identify potential risks and optimize concomitant medications. During treatment, the core task is to monitor and manage irreversible adverse events (irAEs), paying particular attention to pneumonia, colitis, hepatitis, endocrine disorders, and skin reactions. Patients are educated to recognize early symptoms such as fever, cough, diarrhea, and rash and to report them promptly. The pharmacist also reviews the dosing regimen and pre-treatment medications. Regarding medication education, the treatment principle and expected response pattern must be clearly explained, including potential delayed onset of action or pseudoprogression, precautions for intravenous infusion, a home self-monitoring checklist, and timely communication of warning symptoms. The importance of regular follow-up examinations (especially thyroid function tests) is emphasized. Ultimately, by strengthening follow-up and communication, patient compliance, safety, and confidence in treatment are improved.

[0031] The third report emphasizes the institutionalization of patient self-assessment, specifying the frequency of assessments, such as weekly regular assessments. It also clarifies assessment tools, such as self-assessment scales, and methods of recording and reporting, such as filling out and submitting self-assessment forms to healthcare staff. This helps patients understand when and where to conduct self-monitoring, how to record symptom changes, and how to provide feedback to the medical team, ensuring smooth information flow.

[0032] During implementation, a structured database can be created to store the categorized and refined information described above. The database can be either an SQL relational database or a NoSQL database, depending on the information structure and access requirements.

[0033] Example 2: This study compared risk prediction models for immune-related adverse events in lung cancer patients. Data from 313 patients diagnosed with malignant lung cancer and using PD-1 inhibitors, who visited the First Affiliated Hospital of Shihezi University or Shihezi People's Hospital between October 1, 2023, and July 1, 2024, were included. Of these, 117 (37.38%) experienced irAEs after PD-1 inhibitor treatment. The median time to irAEs was 82 days. A statistically significant difference in PFS (progression-free survival) was observed between the irAEs group and the non-irAEs group (P=0.032).

[0034] Statistical Analysis: Univariate and multivariate logistic regression analyses were performed to identify independent risk factors for irAEs. Table 1 shows that univariate analysis results indicated that BMI, pathological type, ECOG, NLR, PNI, and SII were associated with the occurrence of irAEs. Based on the univariate analysis results, factors with P < 0.05 in the univariate analysis were included in the multivariate logistic regression analysis, as shown in Table 2. P < 0.05 was considered statistically significant, and pathological type, ECOG, NLR, PNI, and SII were independent risk factors for the occurrence of irAEs.

[0035] Table 1 Univariate Analysis Table

[0036] Table 2 Multifactor Analysis Table Evaluation Model Establishment: Based on the independent risk factors obtained from logistic regression analysis—pathological type, ECOG, NLR, PNI, and SII—five independent risk factors were used as feature variables, and irAEs were used as predictor variables. Six machine learning algorithms, including SVM, KNN, LR, RF, XGBoost, and NB, were employed to construct irAEs risk prediction models for lung cancer patients. Patient data were automatically and randomly allocated to the training and test sets in a 2:8 ratio. The AUC of the training sets for the six models ranged from 0.84 to 0.87. The performance evaluation metrics of the models are shown in Table 3. The SVM and RF models had an AUC of 0.87, significantly higher than the other models, indicating excellent classification ability and accuracy. (See Table 3 for details.) Figure 1 The SVM model had an F1 score of 0.79, while the RF model had an F1 score of 0.81. This indicates that the RF model achieved a better balance between precision and recall, demonstrating superior performance and stability. To further compare the performance of each prediction model, clinical decision curves and calibration curves were used for analysis. See details below. Figure 3 and Figure 4The clinical decision curve visually reflects the clinical benefit of the model at different thresholds. The results show that the RF model achieves relatively ideal clinical benefits. The calibration curve is used to evaluate the consistency between the model's predicted probability and the actual probability of occurrence. As shown in the figure below, the calibration curve of the RF model is basically distributed around the reference line, further indicating that the model's prediction results have high reliability and accuracy. During internal model testing, the LR model, RF model, and XGBoost model performed well, with an AUC of 0.85. The SVM model showed a significant decline in performance, indicating possible overfitting. See details... Figure 2 Further comparison of F-scores showed that the SVM model scored 2.73, the LR model 0.81, the RF model 0.79, and the XGBoost model 0.75. It was concluded that the LR and RF models effectively balanced precision and recall (see Table 4 for details). Considering the overall performance of each model on the training set, the RF model was deemed to have superior performance and better transferability. Table 3. Predictive performance of irAEs risk prediction models on the training set. Table 4. Predictive performance of the full-level irAEs risk model on the test set. Based on the model performance evaluation results, the RF model was the best predictive model. Patients were classified as low-risk when the RF model's predicted probability was <0.563, and as high-risk when the predicted probability was >0.563. At this risk cutoff value, the PPV was 0.73 and the NPV was 0.90. Here, PV (Positive Predictive Value) is the positive predictive value and NPV (Negative Predictive Value) is the negative predictive value. To facilitate a more convenient and efficient presentation of the model, a nomogram model of all levels of irAEs risk for lung cancer patients was drawn; see [link to details]. Figure 5According to the nomogram model, the scores of each variable under different values ​​are summed, and the total score is the total score. The risk probability of all levels of irAEs can be determined by finding the corresponding risk value on the prediction line at the bottom of the nomogram based on the total score. The central management module mentioned above is responsible for data verification, synchronization, and conflict resolution. It can be designed based on a microservice architecture. Here, a database management system (DBMS) can be used to store and manage data, the application business logic layer (BLL) handles business rules and processes, and the interface presentation layer provides interactive interfaces for doctors, clinical pharmacists, and patients. For the front-end application, customized interfaces can be developed for doctor terminals, clinical pharmacist terminals, and patient terminals, for example, using front-end frameworks such as React or Vue. For back-end services, back-end frameworks such as Node.js or Spring Boot can be used to provide a RESTful API for data exchange between the front-end application and the central management module. For message queue processing, RabbitMQ or Kafka can be used for asynchronous processing of data synchronization requests. During deployment, a server cluster is configured, which can be deployed on a cloud platform, and resources can be dynamically adjusted according to the load. The database server here uses a high-availability database solution, such as a MySQL cluster or the NoSQL database MongoDB, to ensure data reliability and fast access. The configured storage system can use a distributed file system or object storage, such as Ceph or Amazon S3. The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A risk prediction system for immune-related adverse events in lung cancer treatment, characterized in that, include: The information collection module is used to collect patient information from the hospital information system and the patient self-reporting platform; The risk assessment module is used to predict the risk level of patients developing irreversible adverse events (irAEs) based on the collected information and through machine learning models. The report generation module is used to generate three types of reports based on the risk assessment results, for physicians, clinical pharmacists, and patients. The report sending module is used to send three types of reports to the doctor's terminal, the clinical pharmacist's terminal, and the patient's terminal, respectively. The central management module is used to perform data verification, consistency checks, and real-time synchronization of the collected information and generated reports.

2. A method for predicting the risk of immune-related adverse events in lung cancer treatment, characterized in that, Includes the following steps: S1: Collect patient information, including demographic data, tumor characteristics, and laboratory indicators; S2: Establish a nomogram prediction model based on multivariate logistic regression analysis to quantify the weight of each risk factor; S3: Perform consistency checks on the collected patient information and update patient medication data in real time; S4: Use the RF model to predict the probability of irAEs occurring, generate and send three types of reports; S5: When the data in any of the three types of reports changes, the report is updated synchronously and cross-validated.

3. The method for predicting the risk of immune-related adverse events in lung cancer treatment according to claim 2, characterized in that, The demographic data, tumor characteristics, and laboratory indicators in step S1 are as follows: Demographic characteristics: age, sex, BMI, ECOG score, underlying diseases; Tumor-related indicators: pathological type, TNM stage, treatment method; Laboratory test indicators: platelet count, neutrophil count, lymphocyte count, eosinophil count, monocyte count, serum albumin, creatinine, liver function indicators, lactate dehydrogenase, creatine kinase; Derived nutritional inflammation indices: NLR, PLR, MLR, PNI, SII.

4. The method for predicting the risk of immune-related adverse events in lung cancer treatment according to claim 2, characterized in that, The formula for calculating the weights in step S2 is: ； in, They are professions ,job title Relevance to major The weighting coefficients satisfy the condition. .

5. The method for predicting the risk of immune-related adverse events in lung cancer treatment according to claim 2, characterized in that, The consistency check in step S3 includes checking required fields, data types and ranges, patient ID, treatment stage, and follow-up time.

6. The method for predicting the risk of immune-related adverse events in lung cancer treatment according to claim 2, characterized in that, The three types of reports in step S4 are as follows: First report: Sent to the doctor's terminal, including the risk level of irAEs, matching adverse event warning signals, graded treatment procedures and emergency response strategies; The second report is sent to the clinical pharmacist's terminal and includes the pharmaceutical care plan, medication education content, follow-up schedule, and abnormal handling procedures. The third report is sent to the patient's terminal and includes self-monitoring guidelines, lifestyle suggestions, follow-up arrangements, and symptom recording methods.

Images