Survival data fitting method and intelligent system for treating advanced non-squamous non-small cell lung cancer by combining bevacizumab with cis-platinum
Through feature engineering and transfer learning technology, the accuracy and reliability issues of survival data fitting for the treatment of advanced non-squamous non-small cell lung cancer with bevacizumab combined with cisplatin were solved, achieving more efficient data fitting and more accurate clinical decision support.
Patent Information
- Application Number
- CN202510777761.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-11
- Publication Date
- 2025-09-09
AI Technical Summary
Existing technologies make it difficult to fully tap into the multi-dimensional clinical data of bevacizumab combined with cisplatin in the treatment of advanced non-squamous non-small cell lung cancer, resulting in insufficient accuracy and reliability in fitting survival data. Traditional methods lack the utilization of existing knowledge, resulting in low training efficiency and poor generalization ability.
Using advanced feature engineering and transfer learning technologies, a survival data fitting model is constructed through data collection, preprocessing, feature engineering, transfer learning and model training. This includes hybrid statistical screening, machine learning algorithms, heterogeneous data fusion and multimodal feature extraction. It combines pre-training methods of self-supervision and supervised learning to improve the model's fitting accuracy and generalization ability.
It significantly improves the accuracy and reliability of survival data fitting, provides more accurate basis for clinical decision-making, overcomes the limitations of traditional methods, enhances the training efficiency and adaptability of the model, and supports the optimization of clinical treatment plans and patient prognosis assessment.
Smart Images
Figure SMS_1
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of survival data fitting for lung cancer treatment, and specifically to a survival data fitting method and intelligent system for bevacizumab combined with cisplatin in the treatment of advanced non-squamous non-small cell lung cancer. Background Art
[0002] Advanced non-squamous non-small cell lung cancer poses a serious threat to human health. Bevacizumab combined with cisplatin is a commonly used clinical treatment regimen. Accurately evaluating patient survival data under this regimen is of great significance for clinical decision-making.
[0003] Currently, traditional survival analysis methods, such as the Kaplan-Meier method and the Cox proportional hazards model, only utilize some of the patient's clinical characteristics, making it difficult to process multi-dimensional complex data and fully tap into key information, resulting in insufficient accuracy and reliability in survival data fitting. In feature screening, a single statistical or machine learning algorithm cannot comprehensively assess the association between features and survival data, which can easily lead to omission or redundancy of important features. Furthermore, traditional model training lacks the utilization of existing knowledge and starts from scratch, resulting in low training efficiency and poor generalization ability, making it unable to meet the needs of survival data fitting under bevacizumab combined with cisplatin treatment.
[0004] Therefore, innovative survival data fitting methods and systems are urgently needed to improve fitting accuracy with the help of advanced feature engineering and transfer learning techniques, so as to provide better reference for clinical treatment. Summary of the Invention
[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a survival data fitting method and intelligent system for the treatment of advanced non-squamous non-small cell lung cancer with bevacizumab combined with cisplatin, comprising the following steps: Data collection steps: clinical data of patients with advanced non-squamous non-small cell lung cancer treated with bevacizumab combined with cisplatin were collected, including basic patient information, treatment regimen information, survival time data, and survival status data; Data preprocessing steps: clean the collected clinical data, remove missing values and outliers, and standardize the data; Feature engineering steps: Based on preprocessed clinical data, feature selection algorithms are used to screen key features that have a significant impact on survival time and survival status from basic patient information and treatment plan information. At the same time, feature construction methods are used to combine the original features to generate new composite features; Transfer learning step: Input the selected key features and the newly constructed composite features into the pre-trained model, and use the target dataset to fine-tune the parameters of the pre-trained model to obtain an initialized survival analysis model; Model selection step: Based on the initial survival analysis model after transfer learning, select a suitable survival analysis model from a preset survival analysis model library, wherein the preset survival analysis model library includes but is not limited to the Kaplan-Meier model and the Cox proportional hazards model; Model training steps: Input the clinical data after feature engineering into the selected survival analysis model, train the model through the optimization algorithm, and adopt an ensemble learning strategy to combine and train multiple models of the same type with different parameter settings to obtain a trained survival data fitting model; Survival data fitting step: Use the trained survival data fitting model to fit the survival data of patients with advanced non-squamous non-small cell lung cancer under bevacizumab combined with cisplatin treatment, and output the fitting results.
[0006] Preferably, the specific process of the feature engineering step is as follows: Feature screening based on hybrid statistical screening methods to evaluate the association between features and survival data from multiple dimensions; Feature sorting based on machine learning algorithms uses the random forest algorithm to further sort the initially screened key feature sets; Numerical feature combination construction uses mathematical operations to combine key features of numerical type; Categorical feature derivation construction: For categorical key features, logical operations or combinations are performed using the one-hot encoded vectors.
[0007] Preferably, the specific contents of the hybrid statistical screening method are as follows: Construct a joint distribution model of features and survival time: For each feature, whether it is a feature in the patient's basic information or the treatment plan information, use the non-parametric kernel density estimation method to construct the joint probability density function of the feature and the survival time; Calculate the mutual information between features and survival status: Use the mutual information calculation method in information theory to calculate the mutual information value between each feature and the survival status; Estimation of the causal effect of features on survival time: The propensity score matching (PSM) method in causal inference is combined with dual machine learning technology to estimate the causal effect of each feature on survival time; Comprehensive weight screening: In order to integrate the results of the above three steps, different weights are assigned to each evaluation indicator.
[0008] Preferably, the pre-trained model in the transfer learning step includes: Overall model architecture: It adopts a layered hybrid architecture consisting of a bottom-level multimodal data fusion layer, a middle-level attention-enhanced feature extraction layer, and a top-level survival prediction output layer. Multimodal data fusion layer: Considering that patient clinical data contains structured data and unstructured data, this layer adopts a heterogeneous data fusion strategy; Attention-enhanced feature extraction layer: This layer combines the multi-head self-attention mechanism with the graph attention network; Survival prediction output layer: Based on the enhanced features extracted from the middle layer, the output layer adopts a dual-branch structure; Pre-training strategy: adopt a pre-training method that combines self-supervision and supervised learning.
[0009] Preferably, in the data preprocessing step, the survival time data and the survival status data are coded, and the survival status in the survival status data is coded as 0 and the death status is coded as 1.
[0010] Preferably, the optimization algorithm is a gradient descent algorithm or a stochastic gradient descent algorithm.
[0011] Preferably, after the survival data fitting step, a result evaluation step is further included: evaluating the fitting results based on preset evaluation indicators, and the evaluation indicators include but are not limited to mean square error, mean absolute error, and determination coefficient.
[0012] A survival data fitting system for bevacizumab combined with cisplatin in the treatment of advanced non-squamous non-small cell lung cancer, comprising: A data collection module is used to collect clinical data of patients with advanced non-squamous non-small cell lung cancer treated with bevacizumab combined with cisplatin, the clinical data including basic patient information, treatment regimen information, survival time data, and survival status data; The data preprocessing module is used to clean the collected clinical data, remove missing values and outliers, and standardize the data; A model selection module is used to select a suitable survival analysis model from a preset survival analysis model library based on the preprocessed clinical data. The preset survival analysis model library includes but is not limited to the Kaplan-Meier model and the Cox proportional hazards model; The model training module is used to input the pre-processed clinical data into the selected survival analysis model, train the model through the optimization algorithm, and obtain a trained survival data fitting model; The survival data fitting module is used to fit the survival data of patients with advanced non-squamous non-small cell lung cancer under bevacizumab combined with cisplatin treatment using the trained survival data fitting model and output the fitting results.
[0013] Preferably, a result evaluation module is further included for evaluating the fitting results based on preset evaluation indicators, wherein the evaluation indicators include but are not limited to mean square error, mean absolute error, and determination coefficient.
[0014] Preferably, the data preprocessing module performs encoding processing on the survival time data and the survival status data, encoding the survival status in the survival status data as 0 and encoding the death status as 1.
[0015] It has the following beneficial effects: This survival data fitting method and system for bevacizumab combined with cisplatin in the treatment of advanced non-squamous non-small cell lung cancer uses advanced feature engineering technology to fully explore key information in patients' multi-dimensional clinical data, changing the limitation of traditional methods that only use some clinical features. It can more comprehensively process complex data, effectively improve the accuracy and reliability of survival data fitting, and provide a more accurate basis for clinical decision-making. A feature screening method is designed to overcome the shortcomings of single statistical or machine learning algorithms, comprehensively evaluate the relationship between features and survival data from multiple dimensions, avoid the omission of important features and the retention of redundant features, ensure that the features entering the model are more valuable, and thus improve the performance of the entire fitting model. With the help of transfer learning technology, the traditional model training mode from scratch is changed, and existing knowledge is effectively utilized, which not only greatly improves training efficiency, but also significantly enhances the generalization ability of the model, enabling it to better adapt to the complex needs of fitting survival data of advanced non-squamous non-small cell lung cancer under bevacizumab combined with cisplatin treatment, and provide more powerful support for the optimization of clinical treatment plans and patient prognosis assessment. DETAILED DESCRIPTION
[0016] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0017] The present invention provides a technical solution: a survival data fitting method for treating advanced non-squamous non-small cell lung cancer with bevacizumab combined with cisplatin, comprising the following steps: Data collection steps: clinical data of patients with advanced non-squamous non-small cell lung cancer treated with bevacizumab combined with cisplatin were collected, including basic patient information, treatment regimen information, survival time data, and survival status data; Data preprocessing steps: clean the collected clinical data, remove missing values and outliers, standardize the data, encode the survival time data and survival status data, and encode the survival status as 0 and the death status as 1 in the survival status data; Feature engineering steps: Based on the preprocessed clinical data, a feature selection algorithm is used to screen out key features that have a significant impact on survival time and survival status from basic patient information and treatment plan information. At the same time, the original features are combined to generate new composite features through feature construction methods. The specific process is as follows: Feature screening based on a hybrid statistical screening method evaluates the degree of association between features and survival data from multiple dimensions. The specific contents are as follows: Construct a feature-survival time joint distribution model: For each feature, whether it is a feature in the patient's basic information or the treatment plan information, a nonparametric kernel density estimation method is used to construct a joint probability density function of the feature and survival time. Taking the patient's age feature as an example, the age is divided into several intervals. The survival time distribution of the corresponding patients in each interval is statistically analyzed. The kernel function (such as the Gaussian kernel function) is used for smoothing to obtain the joint probability density of age-survival time. This method can more flexibly capture the complex nonlinear relationship between features and survival time. Compared with traditional parametric methods, it is not restricted by specific distribution assumptions and can more accurately reflect the true distribution characteristics of the data. Calculate feature-survival status mutual information: Using the mutual information calculation method in information theory, the mutual information value between each feature and the survival state is calculated. Mutual information can measure the degree of dependence between two random variables. For feature X and survival state Y, the mutual information I(X;Y) calculation formula is:
[0018] Where p(x,y) is the joint probability distribution of feature X and survival status Y, p(x) and p(y) are their marginal probability distributions respectively. The larger the mutual information value, the closer the correlation between the feature and the survival status. By calculating the mutual information between each feature and the survival status, we can preliminarily screen out the feature set with a strong correlation with the survival status. Estimation of the causal effect of feature-survival time: The propensity score matching (PSM) method in causal inference is combined with dual machine learning technology to estimate the causal effect of each feature on survival time. First, the propensity score is calculated based on multiple patient characteristics (including the characteristics to be screened and other covariates). Patients with similar propensity scores are matched to construct a matched sample set. Then, in the matched sample set, a dual machine learning algorithm (such as a combination of random forest and generalized linear model) is used to estimate the survival time difference between the treatment group (with a certain feature value) and the control group (without the feature value). The causal effect estimate of each feature on survival time is obtained. This method can effectively control the influence of confounding factors and more accurately assess the true causal effect of the feature on survival time. Comprehensive weight screening: In order to integrate the results of the above three steps, different weights are assigned to each evaluation indicator. The weights can be determined using the Analytic Hierarchy Process (AHP). The relative importance of each indicator can be determined through expert scoring or data analysis. For example, if the causal effect is considered more critical in feature screening, it can be given a higher weight. The scores of each feature on the three indicators are weighted and summed according to the weight to obtain a comprehensive score for each feature. A comprehensive score threshold is set, and features above the threshold are screened out as key features that have a significant impact on survival time and survival status. Based on the feature ranking of machine learning algorithms, the random forest algorithm is used to further sort the key feature set that has been initially screened. The random forest algorithm evaluates the importance of each feature to the model's prediction of survival data by calculating the average reduced impurity or average reduced accuracy of the features. Features are sorted from high to low according to their importance scores, and a certain proportion of the top-ranked features (e.g., the top 60%) are selected to form the key features for subsequent analysis. Numerical feature combination construction: For numerical key features, mathematical operations are used to combine them. For example, age is multiplied by the total dose of bevacizumab used to construct a new composite feature "age-bevacizumab dose product", which may reflect the different responses of patients of different age groups to specific doses of drugs. Or the cisplatin use cycle is divided by the patient's physical function score to obtain the "cisplatin cycle-function score ratio", which can explore the potential relationship between the treatment cycle and the patient's physical tolerance. These newly constructed composite features are added to the feature set; Categorical feature derivation and construction: For categorical key features, logical operations or combinations are performed using one-hot encoded vectors. For example, if gender is divided into male and female, and pathological subtypes have multiple categories, new binary feature vectors are generated by performing bitwise AND, OR, and other logical operations on the one-hot encoded vectors of gender and pathological subtype. These new features may reveal the survival patterns of patients under specific gender and pathological subtype combinations. These new features are also included in the feature set, completing the entire feature engineering step. Transfer learning steps: Input the selected key features and the newly constructed composite features into the pre-trained model, and use the target dataset to fine-tune the parameters of the pre-trained model to obtain an initialized survival analysis model. The pre-trained model includes: Overall architecture of the model: It adopts a hierarchical hybrid architecture, consisting of a bottom-level multimodal data fusion layer, a middle-level attention-enhanced feature extraction layer, and a top-level survival prediction output layer. The bottom layer is responsible for integrating multi-dimensional patient data, the middle layer strengthens the extraction of key features, and the top layer implements preliminary predictions of survival data. Multimodal data fusion layer: Considering that patient clinical data contains both structured data (such as age and treatment dose) and unstructured data (such as medical records and imaging report summaries), this layer adopts a heterogeneous data fusion strategy. For structured data, an embedding layer is used to map numerical and categorical data into a low-dimensional vector space. For unstructured text data, a pre-trained language model (such as BERT) is used for feature extraction. Subsequently, a fusion module constructed through a gated recurrent unit (GRU) automatically learns the weights of data from different modalities, achieving efficient data fusion and generating a unified feature vector representation. Attention-enhanced feature extraction layer: This layer combines a multi-head self-attention mechanism with a graph attention network. The multi-head self-attention mechanism is used to capture long-distance dependencies between features at different positions in the fused feature vector and explore potential correlation patterns between data. At the same time, each patient feature is considered a node in the graph, and the relationship between features is considered an edge. A feature relationship graph is constructed and processed using a graph attention network. Attention weights are dynamically assigned based on the degree of correlation between nodes, focusing on key features that are more important for survival prediction and enhancing the model's ability to understand complex feature interactions. Survival prediction output layer: Based on the enhanced features extracted from the middle layer, the output layer adopts a two-branch structure. One branch uses the deep SurvivalNet model dedicated to survival analysis to model the probability distribution of survival time and output the patient's survival probability at different time points. The other branch uses traditional classification models (such as logistic regression) to predict the patient's survival status (survival or death). Finally, the output results of the two branches are weighted and fused to obtain the final survival prediction result, providing a richer knowledge foundation for subsequent transfer learning. Pre-training strategy: A pre-training approach combining self-supervision and supervised learning is employed. During the self-supervised learning phase, various data perturbation tasks are designed, such as randomly masking some features and disrupting the order of features, to allow the model to reconstruct the original data and thereby exploit the inherent feature representations of the data. During the supervised learning phase, a large amount of survival data from other lung cancer patients is used to train the model, using the difference between the predicted survival results and the actual survival data as the loss function (e.g., negative log-likelihood loss combined with cross-entropy loss). This allows the model to acquire a certain level of survival prediction capabilities and form transferable knowledge. Model selection step: Based on the initial survival analysis model after transfer learning, select a suitable survival analysis model from a preset survival analysis model library, wherein the preset survival analysis model library includes but is not limited to the Kaplan-Meier model and the Cox proportional hazards model; Model training steps: Input the feature-engineered clinical data into the selected survival analysis model, train the model using the gradient descent algorithm, and adopt an ensemble learning strategy to combine and train multiple models of the same type with different parameter settings to obtain a trained survival data fitting model; Survival data fitting step: Use the trained survival data fitting model to fit the survival data of patients with advanced non-squamous non-small cell lung cancer under bevacizumab combined with cisplatin treatment, and output the fitting results; Result evaluation step: Evaluate the fitting results based on preset evaluation indicators, which include but are not limited to mean square error, mean absolute error, and determination coefficient.
[0019] A survival data fitting system for bevacizumab combined with cisplatin in the treatment of advanced non-squamous non-small cell lung cancer, comprising: A data collection module is used to collect clinical data of patients with advanced non-squamous non-small cell lung cancer treated with bevacizumab combined with cisplatin, the clinical data including basic patient information, treatment regimen information, survival time data, and survival status data; The data preprocessing module is used to clean the collected clinical data, remove missing values and outliers, and standardize the data; A model selection module is used to select a suitable survival analysis model from a preset survival analysis model library based on the preprocessed clinical data. The preset survival analysis model library includes but is not limited to the Kaplan-Meier model and the Cox proportional hazards model; The model training module is used to input the pre-processed clinical data into the selected survival analysis model, train the model through the optimization algorithm, and obtain a trained survival data fitting model; The survival data fitting module is used to fit the survival data of patients with advanced non-squamous non-small cell lung cancer treated with bevacizumab combined with cisplatin using the trained survival data fitting model and output the fitting results; The result evaluation module is used to evaluate the fitting results based on preset evaluation indicators, which include but are not limited to mean square error, mean absolute error, and determination coefficient.
[0020] The data preprocessing module encodes the survival time data and the survival status data, and encodes the survival status in the survival status data as 0 and the death status as 1.
[0021] Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field and related fields without making creative efforts should fall within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described and explained in the present invention shall be implemented in accordance with conventional means in the field unless otherwise specified or limited.
Claims
1. A survival data fitting method for the treatment of advanced non-squamous non-small cell lung cancer with bevacizumab combined with cisplatin, characterized in that: The following steps are involved: Data collection steps: clinical data of patients with advanced non-squamous non-small cell lung cancer treated with bevacizumab combined with cisplatin were collected, including basic patient information, treatment regimen information, survival time data, and survival status data; Data preprocessing steps: clean the collected clinical data, remove missing values and outliers, and standardize the data; Feature engineering steps: Based on preprocessed clinical data, feature selection algorithms are used to screen key features that have a significant impact on survival time and survival status from basic patient information and treatment plan information. At the same time, feature construction methods are used to combine the original features to generate new composite features; Transfer learning step: Input the selected key features and the newly constructed composite features into the pre-trained model, and use the target dataset to fine-tune the parameters of the pre-trained model to obtain an initialized survival analysis model; Model selection step: Based on the initial survival analysis model after transfer learning, select a suitable survival analysis model from a preset survival analysis model library, wherein the preset survival analysis model library includes but is not limited to the Kaplan-Meier model and the Cox proportional hazards model; Model training steps: Input the clinical data after feature engineering into the selected survival analysis model, train the model through the optimization algorithm, and adopt an ensemble learning strategy to combine and train multiple models of the same type with different parameter settings to obtain a trained survival data fitting model; Survival data fitting step: Use the trained survival data fitting model to fit the survival data of patients with advanced non-squamous non-small cell lung cancer under bevacizumab combined with cisplatin treatment, and output the fitting results.
2. The survival data fitting method for treating advanced non-squamous non-small cell lung cancer with bevacizumab combined with cisplatin according to claim 1, characterized in that: The specific process of the feature engineering step is as follows: Feature screening based on hybrid statistical screening methods to evaluate the association between features and survival data from multiple dimensions; Feature sorting based on machine learning algorithms uses the random forest algorithm to further sort the initially screened key feature sets; Numerical feature combination construction uses mathematical operations to combine key features of numerical type; Categorical feature derivation construction: For categorical key features, logical operations or combinations are performed using the one-hot encoded vectors.
3. The survival data fitting method for treating advanced non-squamous non-small cell lung cancer with bevacizumab combined with cisplatin according to claim 2, characterized in that: The specific contents of the hybrid statistical screening method are as follows: Construct a joint distribution model of features and survival time: For each feature, whether it is a feature in the patient's basic information or the treatment plan information, use the non-parametric kernel density estimation method to construct the joint probability density function of the feature and the survival time; Calculate the mutual information between features and survival status: Use the mutual information calculation method in information theory to calculate the mutual information value between each feature and the survival status; Estimation of the causal effect of features on survival time: The propensity score matching (PSM) method in causal inference is combined with dual machine learning technology to estimate the causal effect of each feature on survival time; Comprehensive weight screening: In order to integrate the results of the above three steps, different weights are assigned to each evaluation indicator.
4. The survival data fitting method for treating advanced non-squamous non-small cell lung cancer with bevacizumab combined with cisplatin according to claim 1, characterized in that: The pre-trained model in the transfer learning step includes: Overall model architecture: It adopts a layered hybrid architecture consisting of a bottom-level multimodal data fusion layer, a middle-level attention-enhanced feature extraction layer, and a top-level survival prediction output layer. Multimodal data fusion layer: Considering that patient clinical data contains structured data and unstructured data, this layer adopts a heterogeneous data fusion strategy; Attention-enhanced feature extraction layer: This layer combines the multi-head self-attention mechanism with the graph attention network; Survival prediction output layer: Based on the enhanced features extracted from the middle layer, the output layer adopts a dual-branch structure; Pre-training strategy: adopt a pre-training method that combines self-supervision and supervised learning.
5. The survival data fitting method for treating advanced non-squamous non-small cell lung cancer with bevacizumab combined with cisplatin according to claim 1, characterized in that: In the data preprocessing step, the survival time data and the survival status data are coded, and the survival status in the survival status data is coded as 0 and the death status is coded as 1.
6. The survival data fitting method for treating advanced non-squamous non-small cell lung cancer with bevacizumab combined with cisplatin according to claim 1, characterized in that: The optimization algorithm is a gradient descent algorithm or a stochastic gradient descent algorithm.
7. The survival data fitting method for treating advanced non-squamous non-small cell lung cancer with bevacizumab combined with cisplatin according to claim 1, characterized in that: After the survival data fitting step, a result evaluation step is also included: evaluating the fitting results based on preset evaluation indicators, and the evaluation indicators include but are not limited to mean square error, mean absolute error, and determination coefficient.
8. An intelligent system for fitting survival data of bevacizumab combined with cisplatin in the treatment of advanced non-squamous non-small cell lung cancer, characterized in that: include: A data collection module is used to collect clinical data of patients with advanced non-squamous non-small cell lung cancer treated with bevacizumab combined with cisplatin, the clinical data including basic patient information, treatment regimen information, survival time data, and survival status data; The data preprocessing module is used to clean the collected clinical data, remove missing values and outliers, and standardize the data; A model selection module is used to select a suitable survival analysis model from a preset survival analysis model library based on the preprocessed clinical data. The preset survival analysis model library includes but is not limited to the Kaplan-Meier model and the Cox proportional hazards model; The model training module is used to input the pre-processed clinical data into the selected survival analysis model, train the model through the optimization algorithm, and obtain a trained survival data fitting model; The survival data fitting module is used to fit the survival data of patients with advanced non-squamous non-small cell lung cancer under bevacizumab combined with cisplatin treatment using the trained survival data fitting model and output the fitting results.
9. The intelligent survival data fitting system for treating advanced non-squamous non-small cell lung cancer with bevacizumab combined with cisplatin according to claim 5, characterized in that: It also includes a result evaluation module for evaluating the fitting results based on preset evaluation indicators, which include but are not limited to mean square error, mean absolute error, and determination coefficient.
10. The intelligent survival data fitting system for treating advanced non-squamous non-small cell lung cancer with bevacizumab combined with cisplatin according to claim 5, characterized in that: The data preprocessing module performs encoding processing on the survival time data and the survival status data, and encodes the survival status in the survival status data as 0 and the death status as 1.