Method and device for constructing prognosis prediction model of patient after esophageal cancer neoadjuvant therapy

Abstract

The invention discloses a method and a device for constructing a prognosis prediction model of a patient after esophageal cancer neoadjuvant therapy. The method is applied to the technical field of prediction models. An initial training data set of an esophageal cancer patient is acquired from a cancer database; performing binary classification on the esophageal cancer patients in the initial training data set according to the ten-year survival states, and obtaining a training data set according to variable characteristics of the esophageal cancer patients in the initial training data set after binary classification and the corresponding survival states within the ten-year period; and finally, training the machine learning algorithm model by using the training data set, and constructing a prognosis prediction model for the patient after the neoadjuvant therapy of the esophageal cancer. In the process, a cancer database is used for selecting data, the influence of data heterogeneity on the performance of the model is reduced, the generalization ability of the model is improved, a prediction time window is extended to ten years, and a longer time dimension is provided for clinical decision making.

Description

Technical Field

[0001] The present application relates to the technical field of prediction models, and in particular to a method and device for constructing a prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer. Background Art

[0002] Esophageal cancer is a highly invasive malignant tumor with a poor prognosis. The prognosis of esophageal cancer refers to the prediction of various possible outcomes and their probabilities after the onset of esophageal cancer, including survival and death. Surgery is still the main treatment for esophageal cancer. With the advancement of medical technology, neoadjuvant chemoradiotherapy combined with surgery has become the standard treatment for locally advanced esophageal cancer.

[0003] At present, the prognosis of esophageal cancer patients is mainly evaluated by the clinical staging of patients (TNM staging). TNM staging is a staging system based on tumor size, lymph node metastasis and distant metastasis, which can reflect the malignancy of the tumor and the prognosis of the patient to a certain extent. In recent years, with the rapid development of big data and machine learning technology, researchers have begun to try to use machine learning algorithms to establish prediction models by analyzing patients' clinical data, pathological data, imaging data, etc., in order to improve the accuracy of esophageal cancer prognosis prediction. However, most of the existing machine learning-based prediction models focus on predicting the 5-year survival rate of patients, with a short prediction time window, and the data sets and model structures used are different, lacking a unified standard, making it difficult to compare and promote models. Therefore, there is an urgent need for a tool that can accurately predict the long-term survival of patients in order to provide patients with more accurate individualized treatment. Summary of the invention

[0004] The present application provides a method and device for constructing a prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer. By extending the prediction window to ten years, the long-term survival of patients can be more comprehensively evaluated, and strict inclusion and exclusion criteria are adopted to improve the generalization ability of the model.

[0005] In a first aspect, the present application provides a method for constructing a prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy, the method comprising:

[0006] Obtain an initial training dataset of esophageal cancer patients from the Cancer Database;

[0007] The esophageal cancer patients in the initial training data set are classified into two categories according to the ten-year survival status, and a training data set is obtained according to the variable characteristics of the esophageal cancer patients in the initial training data set after the two-category classification and the corresponding ten-year survival status, wherein the initial training data set after the two-category classification includes patients who survived within the ten-year period and patients who died within the ten-year period;

[0008] The training data set was used to train the machine learning algorithm model and to construct a prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy.

[0009] Optionally, obtain a dataset of esophageal cancer patients from a cancer database, including:

[0010] An initial data set of esophageal cancer patients was obtained from the cancer database according to the inclusion criteria, including: patients with esophageal squamous cell carcinoma or adenocarcinoma, patients with clinical stage T1-3N0-3M0, and patients who had received neoadjuvant chemoradiotherapy and surgery;

[0011] The initial data set was cleaned according to the exclusion criteria to obtain a data set of esophageal cancer patients, wherein the exclusion criteria included: other pathological tumors of the esophagus, distant metastasis of cancer cells, and missing survival data.

[0012] Optionally, when the machine learning algorithm model is a random forest model, the method of training the machine learning algorithm model using the training data set to construct a prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer includes:

[0013] Extracting variable features of esophageal cancer patients in the training data set, wherein the variable features include basic information of patients, characteristics before clinical treatment, characteristics after clinical treatment, and characteristics after clinical treatment;

[0014] The chi-square test method was used to preliminarily screen out the variable characteristics related to the survival status within a ten-year period;

[0015] Set the random seed and build the initial random forest model based on the training data set;

[0016] Inputting relevant variable features into the random forest initial model, calculating the importance of relevant variable features to the random forest initial model, and obtaining important variable features;

[0017] The final variable features are determined from the important variable features according to clinical accessibility, the random seed is reset, and the prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer is constructed based on the final variable features and the training data set.

[0018] Optionally, a prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy is constructed based on the final variable features and the training data set, including:

[0019] Determine the features in the random forest model based on the final variable features;

[0020] Determine the mtry parameter in the random forest model through grid search;

[0021] The number of ntrees in the random forest model is selected by cross-validation method;

[0022] The training data set was input into the random forest model for training to obtain a prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy.

[0023] Optionally, after constructing the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer according to the final variable features and the training data set, the method further includes:

[0024] Obtain the prediction score of each esophageal cancer patient in the training data set output by the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer;

[0025] Setting a score threshold for the likelihood of survival, dividing esophageal cancer patients whose predicted scores are greater than the score threshold into a high likelihood of survival group, and dividing esophageal cancer patients whose predicted scores are less than the score threshold into a low likelihood of survival group;

[0026] The prognostic prediction model for patients after neoadjuvant treatment of esophageal cancer was verified according to the survival status within ten years corresponding to the high survival possibility group and the low survival possibility group.

[0027] Optionally, the method further comprises:

[0028] The self-sampling Bootstrap method and the resampling data set of the training data set were used to perform internal validation of the prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer. The area under the receiver operating curve (AUC) and the Hosmer-Lemeshow goodness-of-fit test were used to evaluate the prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer.

[0029] Optionally, the method further comprises:

[0030] An initial validation dataset of esophageal cancer patients was obtained from a research database;

[0031] The esophageal cancer patients in the initial validation data set are classified into two categories according to the ten-year survival status, and the variable characteristics of the esophageal cancer patients in the initial validation data set after the two-category classification and the corresponding ten-year survival status are used as the validation data set, wherein the initial validation data set after the two-category classification includes patients who survived within the ten-year period and patients who died within the ten-year period;

[0032] The Bootstrap method and validation data set were used to externally validate the prognostic prediction model for patients with esophageal cancer after neoadjuvant therapy, and the area under the receiver operating curve (AUC) and Hosmer-Lemeshow goodness-of-fit test were used to evaluate the prognostic prediction model for patients with esophageal cancer after neoadjuvant therapy.

[0033] In a second aspect, the present application provides a device for constructing a prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy, the device comprising:

[0034] An acquisition unit 301 is used to acquire an initial training data set of esophageal cancer patients from a cancer database;

[0035] The classification unit 302 is used to perform binary classification on the esophageal cancer patients in the initial training data set according to the ten-year survival status, and obtain a training data set according to the variable characteristics of the esophageal cancer patients in the binary-classified initial training data set and the corresponding ten-year survival status, wherein the binary-classified initial training data set includes patients who survived within the ten-year period and patients who died within the ten-year period;

[0036] The construction unit 303 is used to train the machine learning algorithm model using the training data set to construct a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer.

[0037] Optionally, the acquiring unit 301 is specifically configured to:

[0038] An initial data set of esophageal cancer patients was obtained from the cancer database according to the inclusion criteria, including: patients with esophageal squamous cell carcinoma or adenocarcinoma, patients with clinical stage T1-3N0-3M0, and patients who had received neoadjuvant chemoradiotherapy and surgery;

[0039] The initial data set was cleaned according to the exclusion criteria to obtain a data set of esophageal cancer patients, wherein the exclusion criteria included: other pathological tumors of the esophagus, distant metastasis of cancer cells, and missing survival data.

[0040] Optionally, when the machine learning algorithm model is a random forest model, the construction unit 303 includes:

[0041] The extraction subunit is used to extract variable features of esophageal cancer patients in the training data set, wherein the variable features include basic information of the patients, characteristics before clinical treatment, characteristics after clinical treatment, and characteristics after clinical treatment.

[0042] A screening subunit is used to preliminarily screen out variable characteristics related to the survival status within a ten-year period from variable characteristics by using the chi-square test method;

[0043] Construct a subunit to set the random seed and build an initial random forest model based on the training data set;

[0044] A calculation subunit is used to input the relevant variable features into the random forest initial model, calculate the importance of the relevant variable features to the random forest initial model, and obtain the important variable features;

[0045] The subunit is constructed, and is also used to determine the final variable features from the important variable features according to clinical accessibility, reset the random seed, and construct a patient prognosis prediction model after neoadjuvant treatment of esophageal cancer based on the final variable features and the training data set.

[0046] Optionally, when the construction subunit is used to construct a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer according to the final variable characteristics and the training data set, it is specifically used to:

[0047] Determine the features in the random forest model based on the final variable features;

[0048] Determine the mtry parameter in the random forest model through grid search;

[0049] The number of ntrees in the random forest model is selected by cross-validation method;

[0050] The training data set is input into the random forest model for training to obtain a prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy.

[0051] Optionally, the device 300 further includes: a verification unit, specifically configured to:

[0052] Obtain the prediction score of each esophageal cancer patient in the training data set output by the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer;

[0053] Setting a survival probability score threshold, dividing esophageal cancer patients whose predicted scores are greater than the score threshold into a high survival probability group, and dividing esophageal cancer patients whose predicted scores are less than the score threshold into a low survival probability group;

[0054] The prognostic prediction model for patients after neoadjuvant treatment of esophageal cancer was verified according to the survival status within ten years corresponding to the high survival possibility group and the low survival possibility group.

[0055] Optionally, the device 300 further includes: an internal verification unit, specifically configured to:

[0056] The Bootstrap method and the resampling data set of the training data set were used to perform internal validation of the prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy, and the area under the receiver operating curve (AUC) and the Hosmer-Lemeshow goodness-of-fit test were used to evaluate the prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy.

[0057] Optionally, the apparatus 300 further includes: an external verification unit, specifically configured to:

[0058] An initial validation dataset of esophageal cancer patients was obtained from a research database;

[0059] The esophageal cancer patients in the initial validation data set are classified into two categories according to the ten-year survival status, and the variable characteristics of the esophageal cancer patients in the initial validation data set after the two-category classification and the corresponding ten-year survival status are used as the validation data set, wherein the initial validation data set after the two-category classification includes patients who survived within the ten-year period and patients who died within the ten-year period;

[0060] The Bootstrap method and validation data set were used to externally validate the prognostic prediction model for patients with esophageal cancer after neoadjuvant therapy, and the area under the receiver operating curve (AUC) and Hosmer-Lemeshow goodness-of-fit test were used to evaluate the prognostic prediction model for patients with esophageal cancer after neoadjuvant therapy.

[0061] In a third aspect, the present application provides an electronic device, the electronic device comprising a memory and a processor:

[0062] The memory is used to store computer programs;

[0063] The processor is used to execute the method provided by the first aspect according to the computer program.

[0064] In a fourth aspect, the present application further provides a computer-readable storage medium, which is used to store a computer program, and the computer program is used to execute the method provided in the first aspect above.

[0065] It can be seen that this application has the following beneficial effects:

[0066] The present application provides a method and device for constructing a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer. First, an initial training data set of esophageal cancer patients is obtained from a cancer database; the esophageal cancer patients in the initial training data set are classified into two categories according to the ten-year survival status, and a training data set is obtained according to the variable characteristics of the esophageal cancer patients in the initial training data set after the two-category classification and the corresponding survival status within a ten-year period, wherein the initial training data set after the two-category classification includes patients who survived within a ten-year period and patients who died within a ten-year period; the machine learning algorithm model is trained using the training data set to construct a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer. In this process, a cancer database is used to select data, reduce the impact of data heterogeneity on model performance, improve the generalization ability of the model, and extend the prediction time window to ten years, so as to more comprehensively evaluate the long-term survival of patients and provide a longer-term time dimension for clinical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0068] Figure 1 This is a flow chart of an embodiment of a method for constructing a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer in an embodiment of the present application;

[0069] Figure 2This is a flow chart of an embodiment of a method for constructing a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer in an embodiment of the present application;

[0070] Figure 3 This is a schematic diagram of the architecture of a random forest model in a prognosis prediction model for patients with esophageal cancer after neoadjuvant treatment in an embodiment of the present application;

[0071] Figure 4 This is a schematic diagram of the structure of a device for constructing a prognosis prediction model for patients with esophageal cancer after neoadjuvant treatment in an embodiment of the present application;

[0072] Figure 5 This is a schematic diagram of the structure of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0073] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments 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 creative work are within the scope of protection of the present invention.

[0074] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0075] At present, in recent years, with the rapid development of big data and machine learning technology, researchers have begun to try to use machine learning algorithms to establish prediction models by analyzing patients' clinical data, pathological data, imaging data, etc., in order to improve the accuracy of esophageal cancer prognosis prediction. However, when using machine learning algorithms to predict the prognosis of esophageal cancer patients, the following technical defects will occur:

[0076] (1) The prediction time window is short: Most existing prognostic prediction models for esophageal cancer patients focus on the patient's 5-year survival. However, with the promotion of neoadjuvant chemoradiotherapy, 10-year survival has become possible, but there is currently no prediction model for 10-year survival.

[0077] (2) Data heterogeneity: There are differences in the data source, patient characteristics, inclusion and exclusion criteria used in different esophageal cancer prognosis prediction models. Most of them are data from a local area or a large center. Most models include patients who have undergone surgery. However, surgery alone is not the standard treatment for advanced esophageal cancer, which makes it difficult to compare and promote the models.

[0078] (3) Poor model interpretability: Although some complex machine learning models have high prediction accuracy, their internal mechanisms are difficult to explain, which limits their promotion in clinical applications.

[0079] (4) Lack of external validation: Many prognostic prediction models for esophageal cancer patients have only been validated on internal datasets and lack external validation, so their generalization ability is questionable.

[0080] The main technical defects of the above are the short prediction time window and data heterogeneity. In this regard, in the embodiments of the present application, by extending the prediction window to ten years, the long-term survival of patients can be more comprehensively evaluated, and strict inclusion and exclusion criteria are adopted to improve the generalization ability of the model.

[0081] In specific implementation, the method may include, for example: obtaining an initial training data set of esophageal cancer patients from a cancer database; classifying the esophageal cancer patients in the initial training data set into two categories according to the ten-year survival status, and obtaining a training data set based on the variable characteristics of the esophageal cancer patients in the initial training data set after the two-category classification and the corresponding survival status within the ten-year period, wherein the initial training data set after the two-category classification includes patients who survived within the ten-year period and patients who died within the ten-year period; using the training data set to train a machine learning algorithm model to construct a prognosis prediction model for patients after neoadjuvant treatment for esophageal cancer.

[0082] It can be seen that the method provided by the implementation of this application adopts strict inclusion and exclusion criteria, reduces the impact of data heterogeneity on model performance, improves the generalization ability of the model, and extends the prediction time window to ten years, more comprehensively assessing the long-term survival of patients, and providing a longer-term time dimension for clinical decision-making. By quickly and accurately predicting the patient's ten-year survival rate, it helps doctors to adjust treatment plans in a timely manner and improve treatment efficiency.

[0083] In addition, the random forest algorithm is used in the embodiments of the present application. The algorithm has strong interpretability and can clarify which clinical factors have the greatest impact on the patient's ten-year survival through feature importance analysis, providing a new perspective for clinical research. In addition, the present application is not only verified on internal data sets, but also externally verified, further verifying the reliability of the model.

[0084] The patient prognosis prediction model after neoadjuvant therapy for esophageal cancer constructed in the embodiments of the present application can be applied to different scenarios, such as a clinical decision support scenario: doctors can use the patient prognosis prediction model after neoadjuvant therapy for esophageal cancer, combined with the patient's specific situation, to more accurately assess the patient's prognosis, thereby formulating a more personalized treatment plan for the patient; it can also be a translational research and drug development scenario: scientific research institutions and pharmaceutical companies can use the patient prognosis prediction model after neoadjuvant therapy for esophageal cancer to screen out the benefits of chemotherapy and radiotherapy for specific patient subgroups, thereby accelerating the development of new drugs and new radiotherapy technologies. Of course, the patient prognosis prediction model after neoadjuvant therapy for esophageal cancer in this application can be applied not only to the above-mentioned scenarios, and is not limited here.

[0085] To facilitate understanding of the specific implementation of the prognosis prediction model method for patients with esophageal cancer after neoadjuvant treatment provided in the embodiment of the present application, it will be described below with reference to the accompanying drawings.

[0086] It should be noted that the subject implementing the method for constructing a prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer can be the device for constructing a prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer provided in the embodiment of the present application, and the device for constructing a prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer can be carried in an electronic device or a functional module of an electronic device. The electronic device in the embodiment of the present application can be any device capable of implementing the method for constructing a prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer in the embodiment of the present application, for example, it can be an Internet of Things (IoT) device.

[0087] Figure 1 The present invention provides a flowchart of a method for constructing a prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer provided in an embodiment of the present invention. The method can be applied to a device for constructing a prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer. The device for constructing a prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer can be, for example, Figure 3 The apparatus 300 for constructing a prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy is shown, or the apparatus 300 for constructing a prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy can also be integrated into Figure 4 Functional modules in the electronic device 400 are shown.

[0088] like Figure 1 As shown, the method includes the following S101 to S103:

[0089] S101: Obtain an initial training dataset of esophageal cancer patients from a cancer database.

[0090] In order to construct a prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer, it is first necessary to obtain an initial training data set of esophageal cancer patients from a cancer database, and then classify the esophageal cancer patients in the initial training data set according to the ten-year survival status, and obtain a training data set based on the variable characteristics of the esophageal cancer patients in the initial training data set after the binary classification and the corresponding ten-year survival status, and finally use the training data set to train the machine learning algorithm model to construct a prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer. Therefore, in the embodiment of the present application, the initial training data set of esophageal cancer patients is obtained through S101, which makes preliminary preparations for the subsequent acquisition of the training data set.

[0091] As an example, S101 may include: obtaining an initial data set of esophageal cancer patients from a cancer database according to inclusion criteria, the inclusion criteria including: esophageal squamous cell carcinoma patients or adenocarcinoma patients, clinical stage T1-3N0-3M0 patients, and patients who have received neoadjuvant chemoradiotherapy and surgical treatment; then cleaning the initial data set according to exclusion criteria to obtain an initial training data set of esophageal cancer patients, the exclusion criteria including: other esophageal pathological tumors, distant metastasis of cancer cells, and missing survival data.

[0092] Among them, the data in the cancer database within a certain period of time can be determined as the selected data, and the data that meets the inclusion criteria in the selected data will be used as the initial data set for esophageal cancer patients. Among them, the inclusion criteria include patients with esophageal squamous cell carcinoma and adenocarcinoma as the target groups, and the targeted patient groups have a wide coverage and are representative; in addition, since surgery after neoadjuvant chemoradiotherapy is the current standard treatment for esophageal cancer, the inclusion criteria also include patients who have received neoadjuvant chemoradiotherapy and surgery as the target group. After obtaining the initial data set, it is also necessary to set exclusion criteria to exclude some situations that are not related to esophageal cancer or non-existent data to avoid affecting the accuracy of the prediction model.

[0093] In this process, a cancer database was used in this application, and strict inclusion and exclusion criteria were adopted to reduce the impact of data heterogeneity on model performance, which can improve the generalization ability of the model.

[0094] S102: The esophageal cancer patients in the initial training data set are classified into two categories according to the ten-year survival status, and a training data set is obtained according to the variable characteristics of the esophageal cancer patients in the initial training data set after the binarization and the corresponding ten-year survival status, wherein the initial training data set after the binarization includes patients who survived within the ten-year period and patients who died within the ten-year period.

[0095] Since the patients predicted in this application have undergone neoadjuvant therapy and surgical treatment, 10-year survival has become possible after the promotion of neoadjuvant chemoradiotherapy, so the prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer in this application is a prediction model for 10-year survival, so it is necessary to classify the initial training data set according to the ten-year survival status, specifically, the esophageal cancer patients in the initial training data set are classified according to the ten-year survival status, where patients who survive within the ten-year period are defined as ten-year survivors, and patients who die within the ten-year period are defined as non-ten-year survivors. It should be noted that patients who are recorded as alive in the initial training data set but have been followed up for less than ten years need to be reviewed and removed from the cohort.

[0096] S103: Use the training data set to train the machine learning algorithm model and construct a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer.

[0097] In the embodiments of the present application, a variety of machine learning algorithm models can be used to train the training data set, such as random forest, support vector machine, neural network, XGBoost, etc. These algorithms have their own advantages and disadvantages. For example, support vector machine is good at processing small sample data, neural network has strong learning ability, and XGBoost performs well in efficiency and accuracy.

[0098] Among them, since the random forest algorithm has strong anti-overfitting ability and interpretability, it can effectively process high-dimensional data and provide feature importance information. In the embodiment of this application, the random forest model is used as an example of the machine learning algorithm model to specifically illustrate the training process. The following is a detailed process:

[0099] (1) extracting variable features of esophageal cancer patients in the training data set, wherein the variable features include basic information of patients, characteristics before clinical treatment, characteristics after clinical treatment, and characteristics after clinical treatment;

[0100] Among them, the variable characteristics of esophageal cancer patients can be basic information of patients such as diagnosis age, gender, race, insurance status, median family income, region, etc., as well as clinical pre-treatment characteristics, clinical treatment characteristics and clinical post-treatment characteristics, including tumor size, tumor histology, tumor grade, clinical stage and pathological stage after treatment. Patients are summarized as receiving neoadjuvant chemoradiotherapy (NCRT), neoadjuvant chemotherapy (NCT) or other types of neoadjuvant therapy; clinically relevant perioperative variables can also be included, such as radiotherapy dose, chemotherapy regimen, number of positive lymph nodes, number of harvested lymph nodes, lymph node ratio (Lymph node ratio, LNR, defined as the ratio of pathological metastatic LN to total harvested LN), pathological complete response rate, surgical margin, postoperative hospital stay, 30-day readmission and use of adjuvant therapy.

[0101] (2) The chi-square test method was used to preliminarily screen out variable characteristics related to survival status within a ten-year period.

[0102] The threshold value in the chi-square test method can be set to 0.05 to preliminarily screen out variable characteristics related to the survival status within a ten-year period from the variable characteristics.

[0103] (3) Set the random seed and build an initial random forest model based on the training dataset.

[0104] In order to establish a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer in the future, it can be established based on important features. Therefore, it is necessary to first establish a random forest initial model using the training data set, so as to subsequently determine the variable characteristics related to the random forest initial model and improve the prediction accuracy of the model.

[0105] (4) Inputting the relevant variable features into the random forest initial model, calculating the importance of the relevant variable features to the random forest initial model, and obtaining the important variable features;

[0106] In order to calculate the importance of relevant variable characteristics to the random forest initial model, the relevant variable characteristics are directly input into the random forest initial model. The random forest model will calculate the importance of the variable and its statistical p-value. The larger the p-value, the more the variable contributes to the random forest initial model. The important variable characteristics obtained may include lymph node ratio, T stage after treatment, preoperative stage, resident insurance, histological grade, comorbidity score and patient age.

[0107] (5) The final variable features were determined from the important variable features according to clinical accessibility, the random seed was reset, and a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer was constructed based on the training dataset and the final variable features.

[0108] First, since not all variables are available clinically, for example, the economic status of the patient is likely to affect survival, but the definition of wealth varies between countries / regions, and this indicator is difficult to collect and establish a widely applicable model. Therefore, it is necessary to first screen the variables that can be used as a widely applicable model based on clinical accessibility to determine the final variable characteristics, where the final variable characteristics may include: lymph node ratio, post-treatment T stage, preoperative T stage in preoperative staging, preoperative N stage in preoperative staging, histological grade, comorbidity score, and patient age. Secondly, after determining the final variable characteristics, the random seed can be reset to construct a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer based on the training data set and the final variable characteristics.

[0109] The above-mentioned method of constructing a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer based on the training data set and the final variable characteristics may specifically include: determining the features in the random forest model based on the final variable characteristics; determining the mtry parameter in the random forest model by grid search; selecting the number of ntrees in the random forest model by a cross-validation method; and finally inputting the training data set into the random forest model for training to obtain a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer.

[0110] It should be noted that the above method for modifying the features of the machine learning algorithm model can also be to select features using lasso regression, variance threshold method, correlation analysis method, etc. In addition, existing features can also be combined or transformed to generate new features. New features are generated by performing nonlinear transformation on numerical features.

[0111] After obtaining the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer, the model needs to be verified. The specific process may include: inputting the training data set into the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer, and obtaining the prediction score of each esophageal cancer patient in the training data set. Setting a survival score threshold, dividing esophageal cancer patients with a prediction score greater than the score threshold into a high survival possibility group, and dividing esophageal cancer patients with a prediction score less than the score threshold into a low survival possibility group; finally, the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer is verified according to the survival status within a ten-year period corresponding to the high survival possibility group and the low survival possibility group.

[0112] In the embodiment of the present application, two methods are used to verify the effectiveness of the prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy: one is internal verification to determine the effectiveness of the model, and the other is external verification to further determine the effectiveness of the model.

[0113] The process of internally validating the effectiveness of the model may include: using the Bootstrap method and the resampled data set of the training data set to internally validate the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer, and using the area under the receiver operating curve (AUC) and the Hosmer-Lemeshow goodness of fit test to evaluate the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer. The consistency between the prediction score separated by 10% and the proportion of 10-year survivors may also be evaluated.

[0114] The above resampled data set is generated based on the training set data. The traditional training method generally divides a data set into 7:3, with 70% as the training set and 30% as the internal validation set, but this method has been proven to be inferior to resampling.

[0115] Among them, the process of external validation to further determine the effectiveness of the model may include: obtaining an initial validation data set of esophageal cancer patients from a research database; binary classification of esophageal cancer patients in the initial validation data set according to the ten-year survival status, and using the variable characteristics of esophageal cancer patients in the binary initial validation data set and the corresponding ten-year survival status as a validation data set, wherein the binary initial validation data set includes patients who survived within the ten-year period and patients who died within the ten-year period; using the Bootstrap method and the validation data set to externally validate the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer, and using the subject area under the curve AUC and the Hosmer-Lemeshow goodness of fit test to evaluate the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer.

[0116] The above initial validation data set is an independent sample data set different from the internal data source, that is, the internal data is from the National Cancer Database (NCDB), while the initial validation data set is from the CROSS research data. That is, through internal and external validation, the performance of the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer is comprehensively evaluated to ensure the reliability of the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer.

[0117] In addition, the prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer in the embodiment of the present application can also be compared with other models. For example, according to the eighth edition of TNM staging for esophageal cancer, the AUC value of using clinical staging to predict ten-year survivors is 0.52, the AUC value of using post-treatment staging (ypTNM) to predict ten-year survivors is 0.50, and the AUC value of using the traditional logistic regression model to predict ten-year survivors is 0.61. Therefore, in the population of esophageal cancer who have received neoadjuvant chemoradiotherapy combined with surgery, the random forest model used in the embodiment of the present application has a better ability to predict the prognosis of patients, which can better guide clinical decision-making and improve the prognosis of patients.

[0118] It can be seen that the embodiment of the present application uses a cancer database and adopts strict inclusion and exclusion criteria to reduce the impact of data heterogeneity on model performance, improve the generalization ability of the model, and expand the prediction time window to ten years, so as to more comprehensively evaluate the long-term survival of patients and provide a longer-term time dimension for clinical decision-making. By quickly and accurately predicting the ten-year survival rate of patients, it helps doctors to adjust treatment plans in a timely manner and improve treatment efficiency.

[0119] In order to make the method provided in the embodiment of the present application clearer and easier to understand, Figure 2 A specific example of this method is described.

[0120] In the embodiment of the present application, the random forest model is used as the machine learning algorithm model to explain in detail the entire process of constructing a prognosis prediction model for patients after neoadjuvant treatment for esophageal cancer.

[0121] like Figure 2 As shown, the implementation may include:

[0122] S201: Obtain an initial data set of esophageal cancer patients from a cancer database according to an inclusion criterion, and clean the initial data set according to an exclusion criterion to obtain a data set of esophageal cancer patients.

[0123] Firstly, the data from the National Cancer Database (NCDB) from 2004 to 2015 were selected using the inclusion criteria to obtain the initial data set of esophageal cancer patients. In order to ensure the accuracy of the subsequent prediction model, the data that were not within the scope of the study or had missing survival data were cleaned according to the exclusion criteria to obtain the data set of esophageal cancer patients.

[0124] S202: The esophageal cancer patients in the initial training data set are classified into two categories according to the ten-year survival status, and a training data set is obtained according to the variable characteristics of the esophageal cancer patients in the initial training data set after the binarization and the corresponding ten-year survival status, wherein the initial training data set after the binarization includes patients who survived within the ten-year period and patients who died within the ten-year period.

[0125] In this application, the survival prediction time window for esophageal cancer patients is extended to ten years. Compared with the five-year prediction of the traditional model, the long-term prognosis of the patient can be more comprehensively evaluated. Therefore, it is necessary to classify the ten-year survival status in the initial training data set into two categories to obtain a training data set with survival status within a ten-year period.

[0126] S203: Extract variable features of esophageal cancer patients in the training data set, where the variable features include basic information of the patients, pre-clinical treatment features, clinical treatment features, and post-clinical treatment features.

[0127] Patient variables were annotated and unified according to the 2016 version of the NCDB variable dictionary. Demographic characteristics included age at diagnosis, sex, race, insurance status, median household income, region, and Charlson-Deyo score. Tumor characteristics included tumor size, histology, grade, clinical stage, and pathological stage after treatment, and the TNM stage was converted according to the current eighth edition of the staging system. Clinically relevant perioperative variables were included in this application, such as radiotherapy dose, chemotherapy regimen, number of positive lymph nodes, number of harvested lymph nodes, lymph node ratio (Lymph node ratio, LNR, defined as the ratio of pathological metastatic LN to total harvested LN), pathological complete response rate, surgical margin, postoperative hospital stay, 30-day readmission, and use of adjuvant therapy.

[0128] S204: Use the chi-square test method to preliminarily screen out variable characteristics related to survival status within a ten-year period from the variable characteristics.

[0129] The chi-square test method was used to set the threshold value of 0.05 to preliminarily screen the indicators related to the 10-year survival of patients with advanced esophageal cancer and obtain the relevant variable characteristics. In this process, a strict feature selection method was used to screen out the variable characteristics most related to 10-year survival, which improved the interpretability and prediction accuracy of the model.

[0130] S205: Set the random seed and build an initial random forest model based on the training data set.

[0131] In this application, a random forest algorithm is used to construct a prediction model based on big data. The algorithm has strong anti-overfitting ability and interpretability. Therefore, in order to improve the accuracy of the subsequent construction of a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer, a random forest feature importance analysis can be used to screen out the clinical features that have the greatest impact on ten-year survival and improve the prediction accuracy of the model. Therefore, a random forest initial model is first constructed through S205. Figure 3 As described above, a random forest initial model is constructed based on the training data set, wherein the combiner selects the majority of classification results as the final result in the classification problem, and takes the average of multiple regression results as the final result in the regression problem.

[0132] S206: Inputting the relevant variable features into the random forest initial model, calculating the importance of the relevant variable features to the random forest initial model, and obtaining the important variable features.

[0133] S207: Determine final variable features from important variable features based on clinical availability.

[0134] It is necessary to first screen the variables that can be used as a widely applicable model based on clinical accessibility to determine the final variable characteristics, where the final variable characteristics may include: lymph node ratio, post-treatment T stage, preoperative T stage in preoperative staging, preoperative N stage in preoperative staging, histological grade, comorbidity score and patient age.

[0135] S208: Reset the random seed and build a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer based on the final variable characteristics and training data set.

[0136] After obtaining the final variable features, the random seed can be reset to construct a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer. The specific process can be to determine the features in the random forest model according to the final variable features; determine the mtry parameter in the random forest model through grid search, where the mtry parameter determines the number of features randomly selected from all features for each decision tree at each node split, and the final determined mtry is 3; select the number of ntrees in the random forest model through the cross-validation method, where the optimal number of ntrees is selected through the cross-validation method. Generally, the more trees, the higher the stability and accuracy of the model. However, too many trees will increase the computational cost and memory consumption; the final determined ntree is 500; finally, input the training data set into the random forest model for training to obtain a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer.

[0137] S209: Obtain a prediction score for each esophageal cancer patient in a training data set output by a prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer.

[0138] After constructing the prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy, the prediction score for each esophageal cancer patient in the training dataset can be obtained.

[0139] S210: Setting a survival probability score threshold, dividing esophageal cancer patients whose predicted scores are greater than the score threshold into a high survival probability group, and dividing esophageal cancer patients whose predicted scores are less than the score threshold into a low survival probability group, and performing model verification on the prognosis prediction model for esophageal cancer patients after neoadjuvant therapy according to the survival status within a ten-year period corresponding to the high survival probability group and the low survival probability group.

[0140] The score threshold was set to 5, and the patients were divided into a high survival possibility group (score greater than 5 points) and a low survival possibility group (score less than 5 points). There were 3242 patients in the high survival possibility group and 4866 patients in the low survival possibility group. Among them, 14% of the patients in the high survival possibility group eventually achieved 10-year survival, while only less than 1% of the patients in the low survival possibility group achieved 10-year survival. This can verify the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer.

[0141] S211: The Bootstrap method and the resampling data set of the training data set were used to perform internal validation on the prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy, and the AUC and Hosmer-Lemeshow goodness-of-fit test were used to evaluate the prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy.

[0142] In this application, the resampled data of the training data set was used as internal validation data, that is, the internal validation was performed using the Bootstrap method and 500 resampled data sets, and the AUC was used for evaluation, and the AUC value obtained was 0.90. The Hosmer-Lemeshow goodness of fit test was used to evaluate the calibration of the prognostic prediction model for patients after neoadjuvant treatment for esophageal cancer, as well as the consistency between the predicted value separated by 10% and the observed survivor ratio.

[0143] S212: Obtain an initial verification data set of esophageal cancer patients from a research database, and perform classification processing to obtain a verification data set.

[0144] In addition to internal validation to determine the effectiveness of the model, this application also uses external data to further determine the effectiveness of the model. Therefore, the initial validation data set is derived from the CROSS research data, and 165 people are finally included according to the above inclusion and exclusion criteria. Similarly, the esophageal cancer patients in the initial validation data set are classified into two categories according to the ten-year survival status, and the validation data set is obtained based on the variable characteristics of the esophageal cancer patients in the initial validation data set after the two-category classification and the corresponding survival status within the ten-year period. The initial validation data set after the two-category classification includes patients who survived within the ten-year period and patients who died within the ten-year period.

[0145] S213: The Bootstrap method and validation data set were used to externally validate the prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy, and the AUC and Hosmer-Lemeshow goodness-of-fit test were used to evaluate the prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy.

[0146] Finally, the prediction score of each esophageal cancer patient in the validation data set was divided according to the set score threshold of 5 points. There were 99 patients in the high survival possibility group and 66 patients in the low survival possibility group. 52% of the patients in the high survival possibility group eventually achieved 10-year survival, while only 20% of the patients in the low survival possibility group achieved 10-year survival. The external validation was also performed using the Bootstrap method and the validation data set, and the AUC was used for evaluation, with an AUC value of 0.70. The Hosmer-Lemeshow goodness of fit test was used to evaluate the calibration of the prognostic prediction model for patients after neoadjuvant treatment for esophageal cancer, as well as the consistency between the predicted value separated by 10% and the observed survivor proportion.

[0147] This embodiment provides a method for constructing a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer. It uses a national cancer database and adopts strict inclusion and exclusion criteria to reduce the impact of data heterogeneity on model performance, improve the generalization ability of the model, and expand the prediction time window to ten years, so as to more comprehensively evaluate the long-term survival of patients and provide a longer-term time dimension for clinical decision-making. By quickly and accurately predicting the patient's ten-year survival rate, it helps doctors to adjust treatment plans in a timely manner and improve treatment efficiency. In addition, a random forest algorithm is used in the embodiment of the present application. The algorithm has strong interpretability and can clarify which clinical factors have the greatest impact on the patient's ten-year survival through feature importance analysis, providing a new perspective for clinical research. In addition, the present application has not only been verified on internal data sets, but also externally verified, which further verifies the reliability of the model.

[0148] See also Figure 3 The present application embodiment provides a device 300 for constructing a prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy, the device comprising:

[0149] An acquisition unit 301 is used to acquire an initial training data set of esophageal cancer patients from a cancer database;

[0150] The classification unit 302 is used to perform binary classification on the esophageal cancer patients in the initial training data set according to the ten-year survival status, and obtain a training data set according to the variable characteristics of the esophageal cancer patients in the binary-classified initial training data set and the corresponding ten-year survival status, wherein the binary-classified initial training data set includes patients who survived within the ten-year period and patients who died within the ten-year period;

[0151] The construction unit 303 is used to train the machine learning algorithm model using the training data set to construct a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer.

[0152] Optionally, the acquiring unit 301 is specifically configured to:

[0153] An initial data set of esophageal cancer patients was obtained from the cancer database according to the inclusion criteria, including: patients with esophageal squamous cell carcinoma or adenocarcinoma, patients with clinical stage T1-3N0-3M0, and patients who had received neoadjuvant chemoradiotherapy and surgery;

[0154] The initial data set was cleaned according to the exclusion criteria to obtain a data set of esophageal cancer patients, wherein the exclusion criteria included: other pathological tumors of the esophagus, distant metastasis of cancer cells, and missing survival data.

[0155] Optionally, when the machine learning algorithm model is a random forest model, the construction unit 303 includes:

[0156] The extraction subunit is used to extract variable features of esophageal cancer patients in the training data set, wherein the variable features include basic information of the patients, characteristics before clinical treatment, characteristics after clinical treatment, and characteristics after clinical treatment.

[0157] A screening subunit is used to preliminarily screen out variable characteristics related to the survival status within a ten-year period from variable characteristics by using the chi-square test method;

[0158] Construct a subunit to set the random seed and build an initial random forest model based on the training data set;

[0159] A calculation subunit is used to input the relevant variable features into the random forest initial model, calculate the importance of the relevant variable features to the random forest initial model, and obtain the important variable features;

[0160] The subunit is constructed, and is also used to determine the final variable features from the important variable features according to clinical accessibility, reset the random seed, and construct a patient prognosis prediction model after neoadjuvant treatment of esophageal cancer based on the final variable features and the training data set.

[0161] Optionally, when the construction subunit is used to construct a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer according to the final variable characteristics and the training data set, it is specifically used to:

[0162] Determine the features in the random forest model based on the final variable features;

[0163] Determine the mtry parameter in the random forest model through grid search;

[0164] The number of ntrees in the random forest model is selected by cross-validation method;

[0165] The training data set is input into the random forest model for training to obtain a prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy.

[0166] Optionally, the device 300 further includes: a verification unit, specifically configured to:

[0167] Obtain the prediction score of each esophageal cancer patient in the training data set output by the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer;

[0168] Setting a survival probability score threshold, dividing esophageal cancer patients whose predicted scores are greater than the score threshold into a high survival probability group, and dividing esophageal cancer patients whose predicted scores are less than the score threshold into a low survival probability group;

[0169] The prognostic prediction model for patients after neoadjuvant treatment of esophageal cancer was verified according to the survival status within ten years corresponding to the high survival possibility group and the low survival possibility group.

[0170] Optionally, the device 300 further includes: an internal verification unit, specifically configured to:

[0171] The Bootstrap method and the resampling data set of the training data set were used to perform internal validation of the prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy, and the area under the receiver operating curve (AUC) and the Hosmer-Lemeshow goodness-of-fit test were used to evaluate the prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy.

[0172] Optionally, the apparatus 300 further includes: an external verification unit, specifically configured to:

[0173] An initial validation dataset of esophageal cancer patients was obtained from a research database;

[0174] The esophageal cancer patients in the initial validation data set are classified into two categories according to the ten-year survival status, and the variable characteristics of the esophageal cancer patients in the initial validation data set after the two-category classification and the corresponding ten-year survival status are used as the validation data set, wherein the initial validation data set after the two-category classification includes patients who survived within the ten-year period and patients who died within the ten-year period;

[0175] The Bootstrap method and validation data set were used to externally validate the prognostic prediction model for patients with esophageal cancer after neoadjuvant therapy, and the area under the receiver operating curve (AUC) and Hosmer-Lemeshow goodness-of-fit test were used to evaluate the prognostic prediction model for patients with esophageal cancer after neoadjuvant therapy.

[0176] It should be noted that the specific implementation method and effect of the device 300 for constructing the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer can be found in the above Figure 1 or Figure 2 The relevant descriptions in the provided methods will not be repeated here.

[0177] The present application embodiment also provides an electronic device 400, such as Figure 4 As shown, the electronic device 400 includes a memory 401 and a processor 402:

[0178] The memory 401 is used to store computer programs;

[0179] The processor 402 is used to execute the above Figure 1 or Figure 2 The method provided.

[0180] In addition, the present application also provides a computer-readable storage medium, the computer-readable storage medium is used to store a computer program, and the computer program is used to execute Figure 1 or Figure 2 The method provided.

[0181] Through the description of the above implementation methods, it can be known that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a read-only memory (ROM) / RAM, a magnetic disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in each embodiment of the present application or some parts of the embodiments.

[0182] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described above is only schematic, in which the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the goals of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0183] The above description is merely an exemplary embodiment of the present application and is not intended to limit the protection scope of the present application.

Claims

1. A method for constructing a prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy, characterized in that: The method comprises: Obtain an initial training dataset of esophageal cancer patients from the Cancer Database; The esophageal cancer patients in the initial training data set are classified into two categories according to the ten-year survival status, and a training data set is obtained according to the variable characteristics of the esophageal cancer patients in the initial training data set after the two-category classification and the corresponding ten-year survival status, wherein the initial training data set after the two-category classification includes patients who survived within the ten-year period and patients who died within the ten-year period; The training data set is used to train the machine learning algorithm model to construct a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer.

2. The method according to claim 1, characterized in that: The step of obtaining a data set of esophageal cancer patients from a cancer database comprises: An initial data set of esophageal cancer patients was obtained from the cancer database according to the inclusion criteria, wherein the inclusion criteria included: patients with esophageal squamous cell carcinoma or adenocarcinoma, patients with clinical stage T1-3N0-3M0, and patients who had received neoadjuvant chemoradiotherapy and surgery; The initial data set is cleaned according to exclusion criteria to obtain the data set of the esophageal cancer patient, and the exclusion criteria include: other pathological tumors of the esophagus, distant metastasis of cancer cells, and missing survival data.

3. The method according to claim 1, characterized in that: When the machine learning algorithm model is a random forest model, the method of using the training data set to train the machine learning algorithm model to construct a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer includes: Extracting variable features of esophageal cancer patients in the training data set, wherein the variable features include basic information of patients, characteristics before clinical treatment, characteristics after clinical treatment, and characteristics after clinical treatment; The chi-square test method was used to preliminarily screen out the variable characteristics related to the survival status within a ten-year period from the variable characteristics; Setting a random seed and building an initial random forest model based on the training data set; Inputting the relevant variable features into the random forest initial model, calculating the importance of the relevant variable features to the random forest initial model, and obtaining important variable features; The final variable features are determined from the important variable features according to clinical accessibility, the random seed is reset, and the prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer is constructed according to the final variable features and the training data set.

4. The method according to claim 3, characterized in that The step of constructing the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer according to the final variable characteristics and the training data set includes: Determining features in the random forest model according to the final variable features; Determine the mtry parameter in the random forest model by grid search; Selecting the number of ntrees in the random forest model by a cross-validation method; The training data set is input into the random forest model for training to obtain the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer.

5. The method according to claim 3, characterized in that: After constructing the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer according to the final variable features and the training data set, the method further includes: Obtaining a prediction score for each esophageal cancer patient in the training data set output by the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer; Setting a score threshold for the possibility of survival, dividing the esophageal cancer patients whose predicted scores are greater than the score threshold into a high possibility of survival group, and dividing the esophageal cancer patients whose predicted scores are less than the score threshold into a low possibility of survival group; The prognosis prediction model for patients after neoadjuvant treatment for esophageal cancer is verified according to the survival status within a ten-year period corresponding to the high survival possibility group and the low survival possibility group.

6. The method according to claim 1, characterized in that The method further comprises: The self-sampling Bootstrap method and the resampling data set of the training data set were used to perform internal validation on the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer, and the area under the receiver operating curve (AUC) and the Hosmer-Lemeshow goodness-of-fit test were used to evaluate the prognosis prediction model for patients after neoadjuvant therapy for esophageal cancer.

7. The method according to claim 1, characterized in that The method further comprises: An initial validation dataset of esophageal cancer patients was obtained from a research database; The esophageal cancer patients in the initial validation data set are classified into two categories according to the ten-year survival status, and the variable characteristics of the esophageal cancer patients in the initial validation data set after the two-category classification and the corresponding ten-year survival status are used as the validation data set, wherein the initial validation data set after the two-category classification includes patients who survived within the ten-year period and patients who died within the ten-year period; The Bootstrap method and the validation data set were used to externally validate the prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy, and the subject area under the curve (AUC) and the Hosmer-Lemeshow goodness-of-fit test were used to evaluate the prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy.

8. A device for constructing a prognosis prediction model for patients with esophageal cancer after neoadjuvant therapy, characterized in that: The device comprises: An acquisition unit, used for acquiring an initial training data set of esophageal cancer patients from a cancer database; a classification unit, for performing binary classification on the esophageal cancer patients in the initial training data set according to the ten-year survival status, and obtaining a training data set according to the variable characteristics of the esophageal cancer patients in the binary-classified initial training data set and the corresponding ten-year survival status, wherein the binary-classified initial training data set includes patients who survived within the ten-year period and patients who died within the ten-year period; A construction unit is used to train a machine learning algorithm model using the training data set to construct a prognosis prediction model for patients after neoadjuvant treatment of esophageal cancer.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, wherein the processor is used to execute a program stored in the memory to run the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method according to any one of claims 1 to 7.

Images