Tumor prognosis evaluation method and system based on clinical data analysis

Abstract

The present application relates to the technical field of tumor prognosis prediction, and particularly relates to a tumor prognosis evaluation method and system based on clinical data analysis. Newly added clinical data is collected regularly to train incremental decision trees and add them to a random forest ensemble model; for each historical decision tree, based on the statistical stability of the feature distribution of the newly added data, the usability is evaluated, combined with the prediction bias to calculate a prognosis rule change index; the right-censored samples with definite outcomes in the newly added data are used to calculate an asymmetric misleading effect index; the weight of the historical decision tree is decayed and calibrated according to the two indexes, to reduce the interference of rule obsolescence and data defects; the prognosis of the patient is predicted based on the calibrated model. The present application effectively overcomes the interference of clinical practice evolution and data right-censored on the accuracy of prognosis evaluation, ensures that the ensemble model is always close to the latest clinical reality, and significantly improves the accuracy of tumor prognosis evaluation.

Description

Technical Field

[0001] This invention relates to the technical field of tumor prognosis prediction, and specifically to a method and system for tumor prognosis assessment based on clinical data analysis. Background Technology

[0002] Tumor prognostic assessment refers to the process of predicting and evaluating outcomes such as disease recurrence, metastasis, survival time, and quality of life in cancer patients after treatment, by comprehensively considering multiple dimensions of information including clinical characteristics, imaging examinations, pathological indicators, molecular subtyping, and treatment plans, and using statistical models or machine learning algorithms. Conducting tumor prognostic assessment has significant clinical value: first, it provides objective evidence for clinical diagnosis and treatment decisions; second, it helps patients and their families understand the risk of disease progression; and third, it serves as an important evaluation indicator in new drug development and clinical research to screen for superior treatment options.

[0003] Generally, random forest algorithms are used for tumor prognostic assessment based on clinical data. First, historically accumulated clinical data is used as a training set to construct multiple basic decision trees, forming a preliminary random forest prediction model. Subsequently, the system collects newly occurring clinical data at fixed intervals and uses this data to train a new decision tree, which is then added to the existing set of decision trees. During the final prognostic assessment, the system integrates the prediction results from all decision trees to arrive at a final, more robust, and reliable prognostic risk assessment. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention aims to provide a method and system for tumor prognostic assessment based on clinical data analysis. The specific technical solution adopted is as follows: A method for tumor prognostic assessment based on clinical data analysis, comprising: acquiring newly added clinical data within a preset time window; training and generating an incremental decision tree based on the newly added clinical data; and adding the incremental decision tree to a random forest ensemble model composed of historical decision trees; for each historical decision tree, evaluating the usability of the newly added clinical data to the prognostic pattern changes reflected by the historical decision tree based on the statistical stability of the distribution of each feature value in the newly added clinical data within the training set of the historical decision tree, and combining this with the actual prediction of the newly added clinical data... The deviation between the post-outcome and the predicted outcome from the historical decision tree is analyzed to calculate the prognostic pattern change index of the historical decision tree. Emergent samples that were right-censored in the historical decision tree training samples but showed a clear true outcome in the newly added clinical data are identified. By comparing the true outcome of the emergent samples with the censored outcome used during the training of the historical decision tree, the asymmetric misleading effect index of the historical decision tree is calculated. Based on the prognostic pattern change index and the asymmetric misleading effect index, the decision contribution weight of each historical decision tree in the ensemble model is attenuated and calibrated. Based on the random forest ensemble model with the calibrated contribution weights, the clinical data of the target patient are weighted and ensembled for prediction, and the tumor prognostic assessment result is output.

[0005] Furthermore, the method for obtaining the availability of the new data in relation to the prognostic pattern changes reflected in the historical decision tree includes: obtaining the availability according to the availability calculation formula, which is shown below: In the formula, Indicates the first The newly added clinical data for the first The usability of the prognostic pattern changes reflected in the historical decision trees; Indicates the number of features in the newly added clinical data; Indicates the feature sequence number in the newly added clinical data; Indicates the first The first in the historical decision tree Information gain of each feature; Indicates the first The first new clinical data item The feature relative to the first Statistical similarity of the distribution of the training set of each historical decision tree; when the... When the feature is a continuous variable, ,in, Indicates the first The first new clinical data item The value of each feature; Indicates the first The first of all clinical data in the training set of the historical decision tree The average of the features; Indicates the first The first of all clinical data in the training set of the historical decision tree Standard deviation of each feature; This represents an exponential function with the natural constant as its base; when the th... When the feature is a discrete variable or a categorical variable equal to the The frequency of each feature value in the historical training set.

[0006] Furthermore, the method for obtaining the prognostic pattern change index includes: obtaining the prognostic pattern change index according to the calculation formula of the preliminary prognostic pattern change index, the calculation formula of the preliminary prognostic pattern change index is as follows: In the formula, Indicates the first Preliminary prognostic pattern change indicators of a historical decision tree; This indicates the number of new clinical data entries added within the preset time window; Indicates the first The newly added clinical data for the first The usability of the prognostic pattern changes reflected in the historical decision trees; Indicates the first The newly added clinical data includes a true prognostic outcome indicator; Indicates the first The first new clinical data input Predictive prognostic indicators obtained after building historical decision trees; This represents the normalization function for maximum and minimum values; by introducing the unidirectionality of medical prognosis to enhance the true trend, an index of prognostic pattern changes is obtained, and the calculation formula is shown below: In the formula, Indicates the first Prognostic pattern change indicators of a historical decision tree; Indicates the first The time period for obtaining each historical decision tree; Indicates the first The time period for obtaining each historical decision tree; Indicates the first Preliminary prognostic pattern change indicators of a historical decision tree; Indicates the first Preliminary prognostic pattern change indicators of a historical decision tree; This indicates a sign function, meaning that when the value inside the parentheses is greater than 0, the sign function has a value of 1, and when the value inside the parentheses is equal to 0, the sign function has a value of 0. Indicates except the first All historical decision trees except for the first historical decision tree The average value of the corresponding symbolic function.

[0007] Furthermore, the method for obtaining the asymmetric misleading effect index includes: firstly, obtaining a preliminary asymmetric misleading effect index, the calculation formula of which is shown below: In the formula, Indicates the first A preliminary indicator of the asymmetric misleading effect of a historical decision tree; This indicates the number of samples, and the sample set specifically refers to: the number of samples in the first... The original training set of the historical decision trees was in a right-censored state, but the number of samples with clear endpoint events was observed in the newly added clinical data; In the sample set, the first... The true prognostic outcome indicators for each sample; In the sample set, the first... The sample at the th The censored prognostic outcome index values ​​used during the training of the historical decision trees; and when hour, The preliminary asymmetric misleading effect index was optimized by combining the aforementioned prognostic pattern change index to obtain the asymmetric misleading effect index, the calculation formula of which is shown below: In the formula, Indicates the first An indicator of the asymmetric misleading effect of historical decision trees; Indicates the first Prognostic pattern change indicators of a historical decision tree; Indicates the first A preliminary indicator of the asymmetric misleading effect of a historical decision tree.

[0008] Furthermore, based on the prognostic pattern change index and the asymmetric misleading effect index, the decision contribution weight of each historical decision tree in the ensemble model is attenuated and calibrated, including: using the prognostic pattern change index and the asymmetric misleading effect index as a joint penalty term, calculating the weight of the first historical decision tree in the ensemble model. The calibrated decision contribution weights of each historical decision tree: In the formula, Indicates the first The calibrated decision contribution weights of each historical decision tree; Indicates the first The decision contribution weights of each historical decision tree before calibration; A pre-defined penalty factor representing the prognostic pattern of change; Indicates the first Prognostic pattern change indicators of a historical decision tree; The pre-defined penalty factor representing the asymmetric misleading effect index; Indicates the first An indicator of the asymmetric misleading effect of historical decision trees; This represents an exponential function with the natural constant as its base.

[0009] Furthermore, the preset time window is set to one clinical quarter; the tumor prognostic assessment results include the patient's expected survival time, recurrence risk probability, or disease progression score.

[0010] A tumor prognostic assessment system based on clinical data analysis is provided. The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the tumor prognostic assessment method based on clinical data analysis described above.

[0011] This invention offers the following advantages: Through an incremental learning mechanism, it acquires the latest clinical data at fixed intervals to train new decision trees, ensuring the model remains synchronized with clinical practice. For each historical decision tree in the ensemble model, the invention first assesses the usability of new data through the statistical stability of feature distributions, eliminating noise interference and accurately quantifying whether each new data point reflects the changing patterns. Then, based on availability-weighted prediction bias, preliminary variability is calculated, and the direction is corrected using the medical prior of unidirectional improvement in tumor prognosis over time, obtaining reliable indicators of prognostic pattern variability. Simultaneously, the invention utilizes historical right-censored samples from the new data to back-estimate prediction bias caused by incomplete information in the historical model, and uses variability indicators for attribution purification, obtaining an asymmetric misleading effect indicator. Finally, the above two indicators are used as a joint penalty factor, dynamically reducing the ensemble weights of outdated and biased historical decision trees through an exponential decay function, achieving adaptive calibration of the model ensemble weights. This invention effectively overcomes the interference of clinical practice evolution and data right censoring on the accuracy of prognostic assessment, ensuring the ensemble model always closely reflects the latest clinical reality and significantly improving the accuracy of tumor prognostic assessment. Attached Figure Description

[0012] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A flowchart of a tumor prognostic assessment method based on clinical data analysis provided in one embodiment of the present invention; Figure 2This is a block diagram of a tumor prognostic assessment system based on clinical data analysis, provided as an embodiment of the present invention. Detailed Implementation

[0014] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a tumor prognostic assessment method and system based on clinical data analysis proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0015] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0016] The following description, in conjunction with the accompanying drawings, details the specific scheme of a tumor prognostic assessment method based on clinical data analysis provided by this invention.

[0017] Please see Figure 1 This illustrates a tumor prognostic assessment method based on clinical data analysis provided by an embodiment of the present invention. The method includes: step S1: acquiring new clinical data within a preset time window, training and generating an incremental decision tree based on the new clinical data, and adding the incremental decision tree to a random forest ensemble model composed of historical decision trees.

[0018] First, using historically accumulated clinical data as a training set, multiple basic decision trees are constructed to form a preliminary random forest prediction model. The features of the newly added clinical data are selected from multi-dimensional core clinical indicators of cancer patients. In this embodiment of the invention, the dimensions of the core clinical indicators include age, tumor stage, pathological classification, gene testing results, treatment plan, and basic medical history.

[0019] To address the dynamic evolution of clinical practice, the system establishes a fixed update cycle, such as each clinical quarter. At the end of each cycle, the system automatically collects data from newly diagnosed patients who have completed initial follow-up during that period, forming a new clinical dataset. Based on this latest dataset, a new incremental decision tree is trained using the same algorithm framework as the historical decision tree, and then added to the existing random forest ensemble model. If the system's update cycle is too short, the amount of new clinical data may be insufficient, resulting in a lack of statistical representativeness in the trained incremental decision tree, potentially leading to model overfitting or training failure. Conversely, if the system's update cycle is too long, the model updates will lag, failing to capture changes in prognostic patterns caused by new drug launches and guideline updates, leading to model aging. Therefore, preferably, in one embodiment of the present invention, the preset time window is set to one clinical quarter; the tumor prognostic assessment results include the patient's expected survival time, recurrence risk probability, or disease progression score.

[0020] At this point, the ensemble model contains multiple decision trees trained at different times.

[0021] Step S2: For each historical decision tree, based on the statistical stability of the distribution of each feature value in the new clinical data in the training set of the historical decision tree, assess the usability of the new clinical data to the prognostic pattern changes reflected by the historical decision tree, and calculate the prognostic pattern change index of the historical decision tree by combining the deviation between the actual prognostic outcome of the new clinical data and the predicted outcome of the historical decision tree.

[0022] The assessment of prognostic pattern changes must be based on reliable data points to avoid noise interference. This embodiment measures the usability of the data in reflecting pattern changes by evaluating the consistency of the feature distribution between each new clinical data point and the historical decision tree training set.

[0023] Preferably, in one embodiment of the present invention, the method for obtaining the usability of newly added clinical data in relation to the prognostic pattern changes reflected in the historical decision tree includes: obtaining the usability according to the usability calculation formula, which is shown below: In the formula, Indicates the first The newly added clinical data for the first The usability of the prognostic pattern changes reflected in the historical decision trees; Indicates the number of features in the newly added clinical data; This represents the feature value index in the historical decision tree; Indicates the first The first in the historical decision tree The information gain of each feature is used to measure the importance of that feature in the random forest ensemble model; Indicates the first The first new clinical data item The feature relative to the first Statistical similarity of the distribution of the training set of historical decision trees.

[0024] When the When the features are continuous variables, the statistical similarity is calculated using the Gaussian probability density function. ,in, Indicates the first The first new clinical data item The value of each feature; Indicates the first The first of all clinical data in the training set of the historical decision tree The average of the features; Indicates the first The first of all clinical data in the training set of the historical decision tree The standard deviation of each feature indicates that the new clinical data was generated on the [number]th [period]. A larger standard deviation indicates that the data was generated on the [number]th [period]. The closer the value of a feature is to the mean of the historical distribution, the higher its usability. This represents an exponential function with the natural constant as its base.

[0025] When the When the feature is a discrete variable or a categorical variable equal to the The frequency of a feature's value in the historical training set is considered. The higher the frequency, the more familiar the random forest ensemble model is with that feature's value, and the higher its usability.

[0026] After obtaining the availability of each new data point, the prediction bias of the historical decision tree on the current data is further quantified to assess the obsolescence of its patterns.

[0027] Preferably, in one embodiment of the present invention, the method for obtaining the prognostic pattern change index includes: obtaining the prognostic pattern change index according to the calculation formula of the preliminary prognostic pattern change index, the calculation formula of the preliminary prognostic pattern change index is as follows: In the formula, Indicates the first Preliminary prognostic pattern change indicators of a historical decision tree; This indicates the number of new clinical data entries added within the preset time window; Indicates the first The newly added clinical data for the first The usability of the prognostic pattern changes reflected in the historical decision trees; Indicates the first The newly added clinical data includes a true prognostic outcome indicator; Indicates the first The first new clinical data input Predictive prognostic indicators obtained after building historical decision trees; This represents the maximum and minimum value normalization function.

[0028] In this formula, the normalized availability is used as the weight to calculate the weighted average of the normalized prediction bias. The larger the bias, the further the historical model's patterns deviate from the current reality.

[0029] Considering that the prognosis of tumors typically shows a unidirectional improvement trend with medical advancements, we introduce the unidirectional nature of medical prognosis to enhance the true trend and obtain an index of prognostic pattern change. The calculation formula is shown below: In the formula, Indicates the first Prognostic pattern change indicators of a historical decision tree; Indicates the first The time period for obtaining each historical decision tree; Indicates the first The time period for obtaining each historical decision tree; Indicates the first Preliminary prognostic pattern change indicators of a historical decision tree; Indicates the first Preliminary prognostic pattern change indicators of a historical decision tree; This indicates a sign function, meaning that when the value inside the parentheses is greater than 0, the sign function has a value of 1, and when the value inside the parentheses is equal to 0, the sign function has a value of 0. Indicates except the first All historical decision trees except for the first historical decision tree The average value of the corresponding symbolic function.

[0030] In this formula, since the earlier the historical decision tree is acquired, the higher the value of the time period. The smaller the value, the better. Therefore, if the time difference between two historical decision trees has the same sign as the difference in the preliminary prognostic pattern change index, it indicates that the change shows a reasonable direction of change on the time axis. If the value is greater than 0, and the direction of change is abnormal, then the transition is considered to be caused by noise. It equals 0.

[0031] Step S3: Identify the manifested samples that were right-censored in the historical decision tree training samples but whose true outcomes were clearly defined in the newly added clinical data. By comparing the true outcomes of the manifested samples with the censored outcomes used during the training of the historical decision tree, calculate the asymmetric misleading effect index of the historical decision tree.

[0032] Right-censored samples are those in survival analyses where the endpoint event (relapse, death, disease progression) was not observed due to the end of follow-up, loss to follow-up, or termination of the study. For early clinical data used in older decision trees, many patients had short follow-up periods and were "truncated" before the event occurred, resulting in incomplete information during historical model training and thus systematic prediction bias.

[0033] The newly added data provides a longer follow-up period, allowing the true outcomes of some historically censored samples to "reveal." By comparing the predictions of the historical model for these "revealed outcomes" with the actual new follow-up results, the systematic prediction errors caused by the lack of information in the historical model can be directly quantified. Therefore, samples that were right-censored in the historical decision tree training samples but whose true outcomes were clearly revealed in the new clinical data can be identified. By comparing the true outcomes of the revealed samples with the censored outcomes used during the training of the historical decision tree, the asymmetric misleading effect index of the historical decision tree can be calculated.

[0034] Preferably, in one embodiment of the present invention, the method for obtaining the asymmetric misleading effect index includes: firstly, obtaining a preliminary asymmetric misleading effect index, the calculation formula of which is shown below: In the formula, Indicates the first A preliminary indicator of the asymmetric misleading effect of a historical decision tree; This indicates the number of samples; the sample set specifically refers to: the number of samples in the first... The original training set of the historical decision trees was in a right-censored state, but the number of samples with clear endpoint events was observed in the newly added clinical data; In the sample set, the first... The true prognostic outcome indicators for each sample; In the sample set, the first... The sample at the th In this embodiment of the invention, the censored prognostic outcome index value used during the training of the historical decision tree is typically the last follow-up time or the truncation value; and when hour, .

[0035] The preliminary asymmetric misleading effect index reflects the average absolute prediction bias caused by data defects in historical models.

[0036] However, prediction bias may originate from both pattern changes and data defects. Therefore, in this embodiment of the invention, the preliminary asymmetric misleading effect index is optimized by combining the prognostic pattern change index to obtain the asymmetric misleading effect index. The calculation formula is as follows: In the formula, Indicates the first An indicator of the asymmetric misleading effect of historical decision trees; Indicates the first Prognostic pattern change indicators of a historical decision tree; Indicates the first A preliminary indicator of the asymmetric misleading effect of a historical decision tree.

[0037] In this formula, when At lower levels, the bias is primarily attributed to data censoring. At higher levels, the bias includes outdated components, and the penalty for data defects should be appropriately reduced. Therefore, for Perform negative correlation processing to obtain .

[0038] Step S4: Based on the prognostic pattern change index and the asymmetric misleading effect index, the decision contribution weight of each historical decision tree in the ensemble model is attenuated and calibrated. Based on the random forest ensemble model after contribution weight calibration, the clinical data of the target patient is weighted and ensembled for prediction, and the tumor prognosis assessment result is output.

[0039] The prognostic pattern change index and the asymmetric misleading effect index obtained in the above steps are used as a joint penalty term to decay the voting weight of historical decision trees in the ensemble model. The larger the weight of the tree, the higher the proportion of its decision rule in the final prediction. In one embodiment of the present invention, an exponential decay function is used to accelerate the decline of the weight of historical decision trees that are severely outdated and highly misleading.

[0040] Preferably, in one embodiment of the present invention, the decision contribution weight of each historical decision tree in the ensemble model is attenuated and calibrated based on the prognostic pattern change index and the asymmetric misleading effect index, including: using the prognostic pattern change index and the asymmetric misleading effect index as a joint penalty term, calculating the weight of the first historical decision tree in the ensemble model. The calibrated decision contribution weights of each historical decision tree: In the formula, Indicates the first The calibrated decision contribution weights of each historical decision tree; Indicates the first The decision contribution weights of each historical decision tree before calibration; A pre-defined penalty factor representing the prognostic pattern of change; Indicates the first Prognostic pattern change indicators of a historical decision tree; The pre-defined penalty factor representing the asymmetric misleading effect index; Indicates the first An indicator of the asymmetric misleading effect of historical decision trees; This represents an exponential function with the natural constant as its base.

[0041] In one embodiment of the present invention, and The value can be determined by minimizing the prediction error using cross-validation, and is set as a constant between 0.5 and 2.0, without being limited here. It should be noted that cross-validation is a technique well-known to those skilled in the art, and will not be elaborated upon here.

[0042] In one embodiment of this invention, after calibrating the contribution weights of each historical decision tree, all decision trees are constructed into a dynamic random forest model using a weighted ensemble approach. The prognostic prediction results of each tree are weighted and fused using the calibrated decision contribution weights as coefficients to form the final ensemble output. Decision trees with larger weights have a higher proportion of their decision rules and prognostic judgments in the ensemble result, thereby weakening outdated patterns and censoring misleading information, reinforcing the current true prognostic distribution, and ensuring the model output is stable and reliable. The weighted ensemble model is encapsulated and deployed to a tumor prognostic assessment system and connected to a clinical data interface. For subsequently enrolled patients, the system automatically extracts standardized clinical features, inputs them into the model to obtain independent prediction results from each decision tree, and then sums them according to the calibrated weights to output prognostic indicators such as patient recurrence risk and survival probability.

[0043] In summary, the process involves acquiring new clinical data within a preset time window, training incremental decision trees based on this data, and then adding these incremental decision trees to a random forest ensemble model composed of historical decision trees. For each historical decision tree, the usability of the new clinical data in reflecting prognostic patterns is assessed based on the statistical stability of the distribution of each feature value in the new clinical data within the training set of that historical decision tree. Furthermore, the prognostic pattern change index of the historical decision tree is calculated by considering the deviation between the actual prognostic outcome of the new clinical data and the predicted outcome of the historical decision tree. Samples that exhibit right censoring in the historical decision tree training samples but show a clear true outcome in the new clinical data are identified. By comparing the true outcome of these samples with the censored outcome used during the training of the historical decision tree, the asymmetric misleading effect index of the historical decision tree is calculated. Based on the prognostic pattern change index and the asymmetric misleading effect index, the decision contribution weight of each historical decision tree in the ensemble model is attenuated and calibrated. Finally, based on the weighted random forest ensemble model, a weighted ensemble prediction is performed on the clinical data of the target patient, outputting the tumor prognostic assessment result.

[0044] A second objective of one embodiment of the present invention is to provide a tumor prognostic assessment system based on clinical data analysis. This system includes a memory, a processor, and a computer program. The memory stores the corresponding computer program, and the processor runs the corresponding computer program. When the computer program runs on the processor, it can implement the methods described in steps S1-S4, specifically including: a data acquisition module 101, used to acquire newly added clinical data within a preset time window, train and generate an incremental decision tree based on the newly added clinical data, and add the incremental decision tree to a random forest ensemble model composed of historical decision trees; and a decision tree transition analysis module 102, used to, for each historical decision tree, evaluate the change in prognostic patterns reflected by the newly added clinical data on the historical decision tree based on the statistical stability of the distribution of each feature value in the newly added clinical data in the training set of the historical decision tree. The system calculates the prognostic pattern change index of the historical decision tree by considering the availability of the data and the deviation between the actual prognostic outcome of the newly added clinical data and the predicted outcome of the historical decision tree. The asymmetric misleading effect analysis module 103 identifies manifest samples that were right-censored in the training samples of the historical decision tree but showed a clear actual outcome in the newly added clinical data. By comparing the actual outcome of the manifest samples with the censored outcome used during the training of the historical decision tree, the asymmetric misleading effect index of the historical decision tree is calculated. The prognostic assessment module 104, based on the prognostic pattern change index and the asymmetric misleading effect index, performs attenuation calibration on the decision contribution weight of each historical decision tree in the ensemble model. Based on the random forest ensemble model with the calibrated contribution weights, it performs weighted ensemble prediction on the clinical data of the target patient and outputs the tumor prognostic assessment result.

[0045] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0046] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for tumor prognostic assessment based on clinical data analysis, characterized in that, The method includes: acquiring new clinical data within a preset time window; training and generating an incremental decision tree based on the new clinical data; and adding the incremental decision tree to a random forest ensemble model composed of historical decision trees; for each historical decision tree, assessing the usability of the new clinical data to the prognostic pattern changes reflected by the historical decision tree based on the statistical stability of the distribution of each feature value in the new clinical data in the training set of the historical decision tree; calculating the prognostic pattern change index of the historical decision tree by combining the deviation between the actual prognostic outcome of the new clinical data and the predicted outcome of the historical decision tree; identifying manifest samples that are right-censored in the training samples of the historical decision tree but have a clear actual outcome in the new clinical data; calculating the asymmetric misleading effect index of the historical decision tree by comparing the actual outcome of the manifest samples with the censored outcome used during the training of the historical decision tree; adjusting the decision contribution weight of each historical decision tree in the ensemble model by attenuation based on the prognostic pattern change index and the asymmetric misleading effect index; performing weighted ensemble prediction on the clinical data of the target patient based on the random forest ensemble model after contribution weight adjustment; and outputting the tumor prognostic assessment result.

2. The tumor prognostic assessment method based on clinical data analysis according to claim 1, characterized in that, The method for obtaining the usability of new data in relation to the prognostic pattern changes reflected in the historical decision tree includes: obtaining the usability according to the usability calculation formula, which is shown below: In the formula, Indicates the first The newly added clinical data for the first The usability of the prognostic pattern changes reflected by the historical decision trees; Indicates the number of features in the newly added clinical data; Indicates the feature number in the newly added clinical data; Indicates the first The first in the historical decision tree Information gain of each feature; Indicates the first The first new clinical data item The feature relative to the first Statistical similarity of the distribution of the training set of each historical decision tree; when the... When the feature is a continuous variable, ,in, Indicates the first The first new clinical data item The value of each feature; Indicates the first The first of all clinical data in the training set of the historical decision tree The average of the features; Indicates the first The first of all clinical data in the training set of the historical decision tree Standard deviation of each feature; This represents an exponential function with the natural constant as its base; when the th... When the feature is a discrete variable or a categorical variable equal to the The frequency of each feature value in the historical training set.

3. The tumor prognostic assessment method based on clinical data analysis according to claim 1, characterized in that, The method for obtaining the prognostic pattern change index includes: obtaining the prognostic pattern change index according to the calculation formula of the preliminary prognostic pattern change index, the calculation formula of the preliminary prognostic pattern change index is as follows: In the formula, Indicates the first Preliminary prognostic pattern change indicators of a historical decision tree; This indicates the number of new clinical data entries added within the preset time window; Indicates the first The newly added clinical data for the first The usability of the prognostic pattern changes reflected by the historical decision trees; Indicates the first The newly added clinical data includes a true prognostic outcome indicator; Indicates the first The first new clinical data input Predictive prognostic indicators obtained after building historical decision trees; This represents the normalization function for maximum and minimum values; by introducing the unidirectionality of medical prognosis to enhance the true trend, an index of prognostic pattern changes is obtained, and the calculation formula is shown below: In the formula, Indicates the first Prognostic pattern change indicators of a historical decision tree; Indicates the first The time period for obtaining each historical decision tree; Indicates the first The time period for obtaining each historical decision tree; Indicates the first Preliminary prognostic pattern change indicators of a historical decision tree; Indicates the first Preliminary prognostic pattern change indicators of a historical decision tree; This indicates a sign function, meaning that when the value inside the parentheses is greater than 0, the sign function has a value of 1, and when the value inside the parentheses is equal to 0, the sign function has a value of 0. Indicates except the first All historical decision trees except for the first historical decision tree The average value of the corresponding symbolic function.

4. The tumor prognostic assessment method based on clinical data analysis according to claim 1, characterized in that, The method for obtaining the asymmetric misleading effect index includes: firstly, obtaining a preliminary asymmetric misleading effect index, the calculation formula of which is shown below: In the formula, Indicates the first A preliminary indicator of the asymmetric misleading effect of a historical decision tree; This indicates the number of samples, and the sample set specifically refers to: the number of samples in the first... The original training set of the historical decision trees was in a right-censored state, but the number of samples with clear endpoint events was observed in the newly added clinical data; This indicates that in the sample set, the first... The true prognostic outcome indicators for each sample; This indicates that in the sample set, the first... The sample at the th The censored prognostic outcome index values ​​used during the training of the historical decision trees; and when hour, The preliminary asymmetric misleading effect index was optimized by combining the aforementioned prognostic pattern change index to obtain the asymmetric misleading effect index, the calculation formula of which is shown below: In the formula, Indicates the first An indicator of the asymmetric misleading effect of historical decision trees; Indicates the first Prognostic pattern change indicators of a historical decision tree; Indicates the first A preliminary indicator of the asymmetric misleading effect of a historical decision tree.

5. The tumor prognostic assessment method based on clinical data analysis according to claim 1, characterized in that, Based on the prognostic pattern change index and the asymmetric misleading effect index, the decision contribution weight of each historical decision tree in the ensemble model is attenuated and calibrated, including: using the prognostic pattern change index and the asymmetric misleading effect index as a joint penalty term, calculating the weight of the first historical decision tree. The calibrated decision contribution weights of each historical decision tree: In the formula, Indicates the first The calibrated decision contribution weights of each historical decision tree; Indicates the first The decision contribution weights of each historical decision tree before calibration; A pre-defined penalty factor representing the prognostic pattern of change; Indicates the first Prognostic pattern change indicators of a historical decision tree; The pre-defined penalty factor representing the asymmetric misleading effect index; Indicates the first An indicator of the asymmetric misleading effect of historical decision trees; This represents an exponential function with the natural constant as its base.

6. The tumor prognostic assessment method based on clinical data analysis according to claim 1, characterized in that, The preset time window is set to one clinical quarter; the tumor prognostic assessment results include the patient's expected survival time, recurrence risk probability, or disease progression score.

7. A tumor prognostic assessment system based on clinical data analysis, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the tumor prognostic assessment method based on clinical data analysis as described in any one of claims 1 to 6.

Images