Postoperative risk early warning method and system based on machine learning

Abstract

The invention relates to the technical field of postoperative early warning, particularly discloses a postoperative risk early warning method and system based on machine learning, and belongs to the technical field of postoperative early warning. Determining a bad prognosis set of the operation to be pre-warned and bad data of each bad prognosis in the bad prognosis set; determining an influence factor vector of each bad prognosis in the bad prognosis set and an independent prognosis factor set; determining a plurality of preset non-independent factor sets of each bad prognosis in the bad prognosis set; determining a non-independent prognosis factor set of each bad prognosis in the bad prognosis set; and post-operation risk early warning is realized. According to the method, the postoperative risk influence dimension can be comprehensively covered, data association can be accurately captured, intelligent conversion from data to risk prediction is realized, more accurate and comprehensive postoperative risk early warning is provided for clinic, and the scientificity, accuracy and reliability of postoperative risk early warning are improved.

Description

Technical Field

[0001] The present invention relates to the field of postoperative early warning technology, and in particular to a postoperative risk early warning method and system based on machine learning. Background Art

[0002] Historically, postoperative risk assessment has relied on physician experience and a limited number of clinical indicators, such as simple judgments based on patient age and underlying medical conditions, lacking objectivity and comprehensiveness. With the development of artificial intelligence and medical informatization, some risk assessment models based on statistical methods have emerged. However, these models utilize only single-dimensional data and are unable to capture the relationships between complex clinical factors. With the rise of machine learning technology, it has gradually been introduced into the medical field. However, existing postoperative risk warning methods based on machine learning often lack in-depth data mining and scientific classification, and lack sufficient factor screening to meet the clinical demand for accurate risk warnings.

[0003] Therefore, the present invention proposes a postoperative risk warning method and system based on machine learning. Summary of the Invention

[0004] The present invention provides a postoperative risk warning method and system based on machine learning, which obtains and analyzes historical patient data, determines the poor prognosis set of the operation to be warned, and the poor data of each poor prognosis in the poor prognosis set, determines the influencing factor vector and independent prognostic factor set of each poor prognosis in the poor prognosis set, determines multiple preset non-independent factor sets for each poor prognosis in the poor prognosis set, determines the non-independent prognostic factor set of each poor prognosis in the poor prognosis set, and realizes postoperative risk warning based on historical patient data, all poor prognosis independent prognostic factor sets and non-independent prognostic factor sets in the poor prognosis set. It can comprehensively cover the postoperative risk impact dimensions, accurately capture the preoperative-intraoperative-postoperative data association, realize the intelligent transformation from data to risk prediction, provide more accurate and comprehensive postoperative risk warning for clinical practice, and improve the scientificity, accuracy and reliability of postoperative risk warning.

[0005] The present invention provides a postoperative risk warning method based on machine learning, comprising: S1: Obtain historical patient data of multiple patients who have undergone surgeries to be warned, analyze the historical patient data to determine a poor prognosis set for the surgeries to be warned, and poor data for each poor prognosis in the poor prognosis set; S2: Based on historical patient data, the poor prognosis set, and the poor data of each poor prognosis in the poor prognosis set, determine the influencing factor vector of each poor prognosis in the poor prognosis set and the independent prognostic factor set; S3: determining a plurality of preset non-independent factor sets for each poor prognosis in the poor prognosis set based on the influencing factor vector of each poor prognosis in the poor prognosis set and the independent prognosis factor set; S4: determining a set of independent prognostic factors for each poor prognosis in the poor prognosis set based on all preset independent factor sets for each poor prognosis in the poor prognosis set; S5: Achieve postoperative risk warning based on historical patient data, the set of all independent prognostic factors for adverse prognosis in the adverse prognosis set, and the set of non-independent prognostic factors.

[0006] Preferably, a postoperative risk warning method based on machine learning obtains historical patient data of multiple patients who have undergone surgeries to be warned, including: Obtain historical patient sub-data of each patient who underwent a surgery to be warned within a specified time period, wherein the historical patient sub-data includes patient information data, preoperative physiological data, intraoperative operation data, and postoperative physiological data; The historical patient data is determined based on the historical patient sub-data of all patients who underwent the surgery to be warned within a specified time period.

[0007] Preferably, a postoperative risk warning method based on machine learning analyzes historical patient data to determine a set of adverse prognoses for the surgery to be warned, as well as adverse data for each adverse prognosis in the adverse prognosis set, including: Determine whether there is an adverse event in the postoperative physiological data of the historical patient sub-data of each patient in the historical patient data; if so, extract the adverse data from the postoperative physiological data of the historical patient sub-data of each patient in the historical patient data, wherein the adverse data includes at least one or more adverse sub-data, and the adverse sub-data includes the event patient, the adverse event, the adverse event occurrence time, and the adverse event severity; The extracted adverse data of all patients are statistically analyzed to determine the adverse prognosis set of the surgery to be warned, and the adverse data of each adverse prognosis in the adverse prognosis set are determined, wherein the adverse data includes multiple adverse sub-data of the corresponding adverse events.

[0008] Preferably, a postoperative risk warning method based on machine learning, based on historical patient data, a poor prognosis set, and poor data of each poor prognosis in the poor prognosis set, determines the influencing factor vector of each poor prognosis in the poor prognosis set and the independent prognostic factor set, including: Performing feature extraction on each bad sub-data in the bad data of each bad prognosis in the bad prognosis set, and determining a bad feature vector of each bad sub-data in the bad data of each bad prognosis in the bad prognosis set; Based on the event patient in each adverse sub-data in each adverse data of each adverse prognosis in the adverse prognosis set, feature extraction is performed on the patient information data, preoperative physiological data, and intraoperative operation data of the corresponding historical patient sub-data in the historical patient data to determine the influencing factor vector of each adverse prognosis in the adverse prognosis set and the influencing factor value vector of each adverse sub-data in each adverse data of the adverse prognosis in the adverse prognosis set; Determine the first independent prognostic factor subset of each poor prognosis in the poor prognosis set based on the poor feature vectors, influencing factor value vectors, and Cox proportional hazard model of all poor sub-data in the poor prognosis data of each poor prognosis in the poor prognosis set, and record the number of first influencing factors in the first independent prognostic factor subset; Performing Pearson correlation analysis on the bad feature vectors and influencing factor value vectors of all bad sub-data in each bad data of bad prognosis in the bad prognosis set, and determining the single correlation value between each influencing factor in the influencing factor vector and the bad prognosis; Sorting the single correlation values of all influencing factors and poor prognosis in each poor prognosis influencing factor vector in the poor prognosis set from large to small, and selecting the first number of influencing factors after sorting to determine the second independent prognostic factor subset of each poor prognosis in the poor prognosis set; An independent prognostic factor set for each poor prognosis in the poor prognosis set is determined based on the first independent prognostic factor subset and the second independent prognostic factor subset for each poor prognosis in the poor prognosis set, wherein the independent prognostic factor set includes multiple influencing factors.

[0009] Preferably, a postoperative risk warning method based on machine learning, the influencing factor vector of each adverse prognosis in the adverse prognosis set and the independent prognosis factor set, and determining multiple preset non-independent factor sets for each adverse prognosis in the adverse prognosis set, include: Eliminating all influencing factors in the independent prognostic factor set from the influencing factor vector of each poor prognosis in the poor prognosis set, and determining a candidate set of non-independent factors for each poor prognosis in the poor prognosis set, wherein the candidate set of non-independent factors includes multiple influencing factors other than the independent prognostic factor set; Performing cluster analysis on all influencing factors in each candidate set of non-independent factors for poor prognosis in the poor prognosis set, determining multiple cluster sets based on the cluster analysis results, and recording the cluster similarity and the number of second influencing factors of each cluster set; Calculating the number of preset factors for each cluster set based on the cluster similarity of each cluster set of each poor prognosis in the poor prognosis set and the number of second influencing factors; All influencing factors in each cluster set of each poor prognosis in the poor prognosis set are sorted from large to small based on the single correlation value with poor prognosis, and the first preset number of influencing factors after sorting of each cluster set are selected as the preset non-independent factor set of each cluster set; Based on the preset dependent factor sets of all cluster sets for each poor prognosis in the poor prognosis set, multiple preset dependent factor sets for each poor prognosis in the poor prognosis set are determined.

[0010] Preferably, a postoperative risk warning method based on machine learning determines a set of independent prognostic factors for each adverse prognosis in the adverse prognosis set based on all preset independent factor sets for each adverse prognosis in the adverse prognosis set, including: Based on each preset dependent factor set of each poor prognosis in the poor prognosis set and the single correlation values between all influencing factors in each preset dependent factor set and poor prognosis, selecting the optimal subset of each preset dependent factor set of each poor prognosis in the poor prognosis set; The optimal subsets of all preset dependent factor sets for each poor prognosis in the poor prognosis set are not empty sets, and all optimal subsets are extracted. Based on all the extracted optimal subsets, the dependent prognostic factor set for each poor prognosis in the poor prognosis set is determined.

[0011] Preferably, a postoperative risk warning method based on machine learning is implemented based on historical patient data, a set of all independent prognostic factors for poor prognosis in a poor prognosis set, and a set of dependent prognostic factors to achieve postoperative risk warning, including: Based on the set of all independent prognostic factors and non-independent prognostic factors in the poor prognosis set, a postoperative risk classification and early warning model was constructed; The preoperative physiological data and intraoperative operation data of all historical patient sub-data in the historical patient data are used as the input of the postoperative risk classification warning model, and the postoperative physiological data of all historical patient sub-data in the historical patient data are used as the output of the postoperative risk classification warning model to train the postoperative risk classification warning model; Based on the trained postoperative risk classification and warning model, patients' postoperative risk warning can be achieved.

[0012] The present invention provides a postoperative risk warning system based on machine learning, which is used to implement any one of the postoperative risk warning methods based on machine learning in Examples 1 to 7, including: Acquisition module: acquiring historical patient data of multiple patients who have undergone surgeries to be warned, analyzing the historical patient data to determine a poor prognosis set for the surgeries to be warned, and poor data for each poor prognosis in the poor prognosis set; Independent prognosis module: Based on historical patient data, poor prognosis set and poor data of each poor prognosis in the poor prognosis set, determine the influencing factor vector of each poor prognosis in the poor prognosis set and the independent prognostic factor set; Preset non-independent module: based on the influencing factor vector of each poor prognosis in the poor prognosis set and the independent prognosis factor set, determining multiple preset non-independent factor sets for each poor prognosis in the poor prognosis set; Dependent prognosis module: determining a dependent prognostic factor set for each poor prognosis in the poor prognosis set based on all preset dependent factor sets for each poor prognosis in the poor prognosis set; Early warning module: Based on historical patient data, the set of all independent prognostic factors for adverse prognosis in the adverse prognosis set, and the set of non-independent prognostic factors, postoperative risk warning is achieved.

[0013] The beneficial effects of the present invention compared to the prior art are as follows: by acquiring and analyzing historical patient data, determining the adverse prognosis set of the surgery to be warned, and the adverse data of each adverse prognosis in the adverse prognosis set, determining the influencing factor vector and independent prognosis factor set of each adverse prognosis in the adverse prognosis set, determining multiple preset non-independent factor sets for each adverse prognosis in the adverse prognosis set, determining the non-independent prognosis factor set of each adverse prognosis in the adverse prognosis set, and realizing postoperative risk warning based on historical patient data, the independent prognosis factor set of all adverse prognoses in the adverse prognosis set, and the non-independent prognosis factor set. It can comprehensively cover the dimensions of postoperative risk impact, accurately capture the preoperative-intraoperative-postoperative data association, realize the intelligent transformation from data to risk prediction, provide more accurate and comprehensive postoperative risk warning for clinicians, improve the scientificity, accuracy and reliability of postoperative risk warning, promote the construction of high-quality industry data sets in the medical field, and lay the foundation for building large-scale model data products for the medical industry.

[0014] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in this application document.

[0015] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is a flow chart of a postoperative risk warning method based on machine learning in an embodiment of the present invention; Figure 2 Schematic diagram of a postoperative risk warning system based on machine learning in an embodiment of the present invention. Specific reality Implementation method The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention. DETAILED DESCRIPTION

[0017] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention. Example 1

[0018] The present invention provides a postoperative risk warning method based on machine learning, referring to Figure 1 ,include: S1: Obtain historical patient data of multiple patients who have undergone surgeries to be warned, analyze the historical patient data to determine a poor prognosis set for the surgeries to be warned, and poor data for each poor prognosis in the poor prognosis set; S2: Based on historical patient data, the poor prognosis set, and the poor data of each poor prognosis in the poor prognosis set, determine the influencing factor vector of each poor prognosis in the poor prognosis set and the independent prognostic factor set; S3: determining a plurality of preset non-independent factor sets for each poor prognosis in the poor prognosis set based on the influencing factor vector of each poor prognosis in the poor prognosis set and the independent prognosis factor set; S4: determining a set of independent prognostic factors for each poor prognosis in the poor prognosis set based on all preset independent factor sets for each poor prognosis in the poor prognosis set; S5: Achieve postoperative risk warning based on historical patient data, the set of all independent prognostic factors for adverse prognosis in the adverse prognosis set, and the set of non-independent prognostic factors.

[0019] In this example, historical data from multiple patients who underwent surgeries for which a warning is needed is first collected from the medical system. This data includes patient information, pre- and post-operative physiological data, and intraoperative procedural data. Post-operative physiological data is analyzed to determine whether adverse events (such as infection and bleeding) have occurred. If so, adverse data is extracted, including information such as the patient, adverse event, time of occurrence, and severity. Statistically summarized, a set of adverse prognoses and corresponding adverse data is generated, laying the foundation for subsequent analysis.

[0020] In this example, features are extracted from the negative sub-data within the negative data to form a negative feature vector. Based on the event patient, patient information and preoperative and intraoperative data features are extracted from historical patient data to determine the influencing factor vector and influencing factor value vector. Using the Cox proportional hazards model and Pearson correlation analysis, a set of independent prognostic factors with a significant impact on poor prognosis is screened.

[0021] In this embodiment, independent prognostic factors are removed from the influencing factor vector to obtain a candidate set of non-independent factors. Cluster analysis is performed on the candidate set of non-independent factors, and a preset number of factors is calculated based on cluster similarity and the number of factors. The factors are then sorted based on their single correlation values with poor prognosis, and the pre-set number of factors are selected to form the preset set of non-independent factors for each cluster.

[0022] In this embodiment, based on the single correlation value between each factor in the preset dependent factor set and poor prognosis, the optimal subset of each set is selected, and the non-empty optimal subsets are extracted and integrated to form a dependent prognostic factor set.

[0023] In this embodiment, historical patient data, an independent prognostic factor set, and a non-independent prognostic factor set are used to construct a postoperative risk warning model, which is trained with preoperative and intraoperative data as input and postoperative data as output, and ultimately, a postoperative risk warning for patients is achieved based on the trained model.

[0024] The beneficial effects of the above technology are: by acquiring and analyzing historical patient data, determining the adverse prognosis set of surgeries to be warned, as well as the adverse data of each adverse prognosis in the adverse prognosis set, determining the influencing factor vector and independent prognostic factor set of each adverse prognosis in the adverse prognosis set, determining multiple preset non-independent factor sets for each adverse prognosis in the adverse prognosis set, determining the non-independent prognostic factor set for each adverse prognosis in the adverse prognosis set, and realizing postoperative risk warning based on historical patient data, the independent prognostic factor set of all adverse prognoses in the adverse prognosis set, and the non-independent prognostic factor set. It can comprehensively cover the dimensions of postoperative risk impact, accurately capture the preoperative, intraoperative, and postoperative data associations, realize the intelligent transformation from data to risk prediction, provide more accurate and comprehensive postoperative risk warnings for clinicians, and improve the scientificity, accuracy, and reliability of postoperative risk warnings. Example 2

[0025] Based on Example 1, a postoperative risk warning method based on machine learning obtains historical patient data of multiple patients who have undergone surgeries to be warned, including: Obtain historical patient sub-data of each patient who underwent a surgery to be warned within a specified time period, wherein the historical patient sub-data includes patient information data, preoperative physiological data, intraoperative operation data, and postoperative physiological data; The historical patient data is determined based on the historical patient sub-data of all patients who underwent the surgery to be warned within a specified time period.

[0026] In this example, the timeframe for data collection is clearly defined to avoid data interference caused by changes in medical technology and treatment standards over a long time span, ensuring data homogeneity and timeliness. For example, data from the past three years is selected, as surgical techniques and anesthesia methods at that time are more consistent with the current medical environment and can effectively reflect the actual situation of current surgeries.

[0027] In this embodiment, multi-dimensional data is collected for each patient who undergoes a surgery to be warned. Patient information data includes age, gender, underlying diseases, family medical history, etc., which can reflect the patient's individual characteristics; preoperative physiological data includes blood routine, electrocardiogram, liver and kidney function indicators, etc., reflecting the patient's physical function before surgery; intraoperative operation data records the operation duration, blood loss, anesthesia method, surgical instrument use, etc., presenting details of the surgical process; postoperative physiological data includes vital signs, wound healing status, complications, etc., showing the postoperative recovery status.

[0028] In this example, the historical patient sub-data for all patients within a specified time period are aggregated and integrated to form a complete historical patient dataset. This process systematizes the scattered individual data and builds a structured database, providing a comprehensive and unified data foundation for subsequent analysis and model training.

[0029] The beneficial effects of the above technology are: obtaining historical patient data of multiple patients who have undergone surgeries to be warned can improve data integrity and provide a data basis for determining the poor prognosis set and the poor data of each poor prognosis in the poor prognosis set. Example 3

[0030] Based on Example 2, a postoperative risk warning method based on machine learning analyzes historical patient data to determine a set of adverse prognoses for surgeries to be warned, as well as adverse data for each adverse prognosis in the adverse prognosis set, including: Determine whether there is an adverse event in the postoperative physiological data of the historical patient sub-data of each patient in the historical patient data; if so, extract the adverse data from the postoperative physiological data of the historical patient sub-data of each patient in the historical patient data, wherein the adverse data includes at least one or more adverse sub-data, and the adverse sub-data includes the event patient, the adverse event, the adverse event occurrence time, and the adverse event severity; The extracted adverse data of all patients are statistically analyzed to determine the adverse prognosis set of the surgery to be warned, and the adverse data of each adverse prognosis in the adverse prognosis set are determined, wherein the adverse data includes multiple adverse sub-data of the corresponding adverse events.

[0031] In this example, each patient's postoperative physiological data within their historical patient sub-data is first reviewed individually. This postoperative physiological data includes a wide range of key information, such as vital signs (temperature, pulse, respiration, blood pressure, etc.), laboratory test indicators (blood routine, blood biochemistry, coagulation function, etc.), and the functional status of various body systems. This data reflects the patient's actual physical condition after surgery.

[0032] In this embodiment, the review process determines whether there are adverse events. Adverse events are those that negatively impact the patient's postoperative recovery, such as postoperative infection (such as wound infection and lung infection), bleeding (surgical site bleeding and internal bleeding), organ dysfunction (such as heart failure and renal failure), thrombosis (deep vein thrombosis and pulmonary embolism), and neurological complications (such as cognitive impairment and nerve damage).

[0033] In this embodiment, once an adverse event is detected, relevant adverse data is extracted from the patient's postoperative physiological data. Adverse data consists of multiple adverse sub-data. The "event patient" specifies the specific patient involved in the adverse event, facilitating subsequent tracking and analysis of individual conditions; the adverse event details the specific adverse condition that occurred, the "adverse onset time" records the time point when the adverse event began to occur, and the "adverse degree" quantifies the severity of the adverse event, such as the severity of the infection, the specific degree of organ damage, and the number of complications, in order to measure its impact on the patient's health.

[0034] In this embodiment, adverse data extracted from all patients are collected and summarized. These adverse data may come from different patients and cover various types of adverse events and corresponding detailed information.

[0035] In this embodiment, statistical analysis is performed on the aggregated adverse data. By analyzing the type, frequency, and severity of adverse events, similar adverse events are categorized and organized to determine a set of adverse outcomes for the surgeries to be alerted. The adverse outcome set is a collection of various adverse outcomes that may occur in this type of surgery.

[0036] In this embodiment, for each adverse prognosis in the adverse prognosis set, the corresponding adverse data is further clarified. For each specific adverse prognosis, multiple related adverse sub-data are recorded in detail. These data can more meticulously describe the characteristics and circumstances of the adverse prognosis, including the patients involved, the specific manifestations of the adverse event, the time of occurrence, and the severity.

[0037] The beneficial effects of the above technology are: analyzing historical patient data to determine the adverse prognosis set of surgeries to be warned, as well as the adverse data of each adverse prognosis in the adverse prognosis set, which can lay a solid data foundation for subsequent in-depth analysis of the influencing factors of adverse prognosis and the realization of accurate postoperative risk warning. Example 4

[0038] Based on Example 3, a postoperative risk warning method based on machine learning is provided, which determines the influencing factor vector of each poor prognosis in the poor prognosis set and the independent prognostic factor set based on historical patient data, a poor prognosis set, and poor data of each poor prognosis in the poor prognosis set, including: Performing feature extraction on each bad sub-data in the bad data of each bad prognosis in the bad prognosis set, and determining a bad feature vector of each bad sub-data in the bad data of each bad prognosis in the bad prognosis set; Based on the event patient in each adverse sub-data in each adverse data of each adverse prognosis in the adverse prognosis set, feature extraction is performed on the patient information data, preoperative physiological data, and intraoperative operation data of the corresponding historical patient sub-data in the historical patient data to determine the influencing factor vector of each adverse prognosis in the adverse prognosis set and the influencing factor value vector of each adverse sub-data in each adverse data of the adverse prognosis in the adverse prognosis set; Determine the first independent prognostic factor subset of each poor prognosis in the poor prognosis set based on the poor feature vectors, influencing factor value vectors, and Cox proportional hazard model of all poor sub-data in the poor prognosis data of each poor prognosis in the poor prognosis set, and record the number of first influencing factors in the first independent prognostic factor subset; Performing Pearson correlation analysis on the bad feature vectors and influencing factor value vectors of all bad sub-data in each bad data of bad prognosis in the bad prognosis set, and determining the single correlation value between each influencing factor in the influencing factor vector and the bad prognosis; Sorting the single correlation values of all influencing factors and poor prognosis in each poor prognosis influencing factor vector in the poor prognosis set from large to small, and selecting the first number of influencing factors after sorting to determine the second independent prognostic factor subset of each poor prognosis in the poor prognosis set; An independent prognostic factor set for each poor prognosis in the poor prognosis set is determined based on the first independent prognostic factor subset and the second independent prognostic factor subset for each poor prognosis in the poor prognosis set, wherein the independent prognostic factor set includes multiple influencing factors.

[0039] In this embodiment, feature extraction is performed on each sub-data element within each adverse data element within the adverse prognosis set. For example, features such as the event type (e.g., infection, bleeding), nature (acute or chronic), specific numerical values of the adverse event, and time interval are extracted. In this manner, a feature vector is determined for each sub-data element. This vector contains key feature information for that sub-data element and is used for subsequent analysis.

[0040] In this embodiment, based on the event patient in each adverse sub-data in each adverse data of each adverse prognosis in the adverse prognosis set, the corresponding historical patient sub-data is found in the historical patient data. Feature extraction is performed on the patient information data (such as age, gender, underlying diseases, etc.), preoperative physiological data (such as various physiological indicators, examination results, etc.), and intraoperative operation data (such as surgical method, operation duration, blood loss, etc.) in these corresponding historical patient sub-data. After feature extraction, an influencing factor vector is determined for each adverse prognosis in the adverse prognosis set. This vector contains various influencing factors that may be related to the adverse prognosis. At the same time, for each adverse sub-data, its corresponding influencing factor value vector is determined. This vector represents the specific value or degree of influence of each influencing factor in the case of the adverse sub-data.

[0041] In this embodiment, the adverse feature vectors and influencing factor value vectors of all adverse sub-data in the adverse data of each adverse prognosis in the adverse prognosis set are comprehensively considered, and the Cox proportional hazard model is used. Through analysis and calculation of the model, it is determined which factors have a significant independent impact on the adverse prognosis, thereby obtaining the first independent prognostic factor subset for each adverse prognosis in the adverse prognosis set, and the number of influencing factors in this subset, that is, the first influencing factor number, is recorded.

[0042] In this embodiment, the Cox proportional hazards model is a statistical model used to analyze survival data, and is used here to assess the impact of various factors on the risk of poor prognosis.

[0043] In this embodiment, a Pearson correlation analysis is performed on the bad feature vectors and influencing factor value vectors of all bad sub-data in each bad data with a bad prognosis in the bad prognosis set.

[0044] In this embodiment, the Pearson correlation coefficient is an indicator that measures the degree of linear correlation between two variables, and its value range is between -1 and 1. Through this analysis, the single correlation value between each influencing factor in the influencing factor vector and poor prognosis is determined to measure the degree of linear association between each influencing factor and poor prognosis.

[0045] In this embodiment, all the single correlation values of the influencing factors and the poor prognosis in the influencing factor vector of each poor prognosis in the poor prognosis set are sorted in descending order, and according to the number of the first influencing factors recorded previously, the corresponding number of influencing factors at the top of the sorting are selected. These factors constitute the second independent prognostic factor subset of each poor prognosis in the poor prognosis set.

[0046] In this embodiment, the first independent prognostic factor subset and the second independent prognostic factor subset of each poor prognosis in the poor prognosis set are integrated. The integrated set is the independent prognostic factor set of each poor prognosis in the poor prognosis set. This set includes multiple factors that have a significant independent impact on the poor prognosis.

[0047] The beneficial effects of the above technology are: based on historical patient data, poor prognosis sets and poor data of each poor prognosis in the poor prognosis set, the influencing factor vector of each poor prognosis in the poor prognosis set and the independent prognostic factor set are determined, which can comprehensively explore the key independent factors affecting poor prognosis and avoid missing factors, providing a more accurate and reliable basis for postoperative risk warning, and improving the accuracy and effectiveness of risk assessment. Example 5

[0048] Based on Example 4, a postoperative risk warning method based on machine learning determines multiple preset dependent factor sets for each adverse prognosis in the adverse prognosis set based on the influencing factor vector of each adverse prognosis in the adverse prognosis set and the independent prognosis factor set, including: Eliminating all influencing factors in the independent prognostic factor set from the influencing factor vector of each poor prognosis in the poor prognosis set, and determining a candidate set of non-independent factors for each poor prognosis in the poor prognosis set, wherein the candidate set of non-independent factors includes multiple influencing factors other than the independent prognostic factor set; Performing cluster analysis on all influencing factors in each candidate set of non-independent factors for poor prognosis in the poor prognosis set, determining multiple cluster sets based on the cluster analysis results, and recording the cluster similarity and the number of second influencing factors of each cluster set; Calculating the number of preset factors for each cluster set based on the cluster similarity of each cluster set of each poor prognosis in the poor prognosis set and the number of second influencing factors; All influencing factors in each cluster set of each poor prognosis in the poor prognosis set are sorted from large to small based on the single correlation value with poor prognosis, and the first preset number of influencing factors after sorting of each cluster set are selected as the preset non-independent factor set of each cluster set; Based on the preset dependent factor sets of all cluster sets for each poor prognosis in the poor prognosis set, multiple preset dependent factor sets for each poor prognosis in the poor prognosis set are determined.

[0049] In this embodiment, the influencing factor vector of each poor prognosis in the known poor prognosis set covers all factors that may affect the poor prognosis, and the independent prognostic factor set contains key factors that have a direct and independent impact on poor prognosis. On this basis, all factors in the independent prognostic factor set are removed from the influencing factor vector. For example, if age and underlying diseases are independent prognostic factors, they are removed from the influencing factor vector, and the remaining factors constitute the non-independent factor candidate set, focusing on non-independent influencing factors that may have synergistic effects.

[0050] In this embodiment, all influencing factors in the candidate set of non-independent factors are clustered and analyzed, such as by using K-means, hierarchical clustering and other algorithms. Factors are divided into different cluster sets based on the similarity of data features between factors, clinical relevance, etc. For example, factors related to recovery from surgical trauma (such as the size of surgical incision, the degree of tissue damage during surgery) are clustered into one category. The cluster set may be a cluster set based on physiological function, which includes influencing factors such as heart rate and abnormal electrocardiogram indicators, or it may be a cluster set based on disease risk, which includes influencing factors such as hyperlipidemia and hyperglycemia. At the same time, the cluster similarity of each cluster set is recorded to measure the closeness of the correlation between factors within the class; the number of second influencing factors, that is, the total number of factors in each cluster set, is recorded to provide a basis for subsequent screening.

[0051] In this embodiment, based on the cluster similarity of each cluster set and the number of second influencing factors, the calculation formula for calculating the number of preset factors in each cluster set can be expressed as: ; in, represents the number of preset factors for the bth cluster set with the ath poor prognosis, Indicates the similarity between the i-th influencing factor and the j-th influencing factor in the b-th cluster set of the a-th poor prognosis, represents the number of second influencing factors in the bth cluster set of the ath poor prognosis, represents the cluster similarity of the bth cluster set with the ath poor prognosis, represents the similarity standard deviation of the b-th cluster set with the a-th poor prognosis, w1 represents the similarity coefficient, β represents the discrete penalty coefficient, γ represents the scale adjustment coefficient, Indicates rounding up.

[0052] In this embodiment, the similarity coefficient w1 ranges from 0 to 1 and may be 0.2, the discrete penalty coefficient β ranges from 0 to 1 and may be 1.5, and the scale adjustment coefficient γ ranges from 0 to 1 and may be 0.8.

[0053] In this embodiment, all influencing factors in each cluster set are sorted from largest to smallest based on their single correlation values with poor prognosis (obtained through Pearson correlation analysis). Based on the calculated number of preset factors, the top-ranked factors are selected to form the preset non-independent factor set for each cluster set.

[0054] In this embodiment, the preset dependent factor sets of all cluster sets under the same poor prognosis are aggregated to form multiple preset dependent factor sets for each poor prognosis in the poor prognosis set. Each preset dependent factor set represents a class of similar dependent factor combinations.

[0055] In this embodiment, the order of all influencing factors in the preset dependent factor set is still sorted according to the corresponding single correlation sizes.

[0056] In this embodiment, the preset dependent factor set of each cluster set of each poor prognosis in the poor prognosis set corresponds to one preset dependent factor set.

[0057] The beneficial effects of the above technology are: based on the influencing factor vector of each adverse prognosis in the adverse prognosis set and the independent prognostic factor set, multiple preset non-independent factor sets for each adverse prognosis in the adverse prognosis set are determined, which can more comprehensively capture the synergistic effects of non-independent factors, provide more detailed and diverse data support for postoperative risk warning, and significantly improve the comprehensiveness and accuracy of risk assessment. Example 6

[0058] On the basis of Example 4, a postoperative risk warning method based on machine learning determines a set of dependent prognostic factors for each poor prognosis in the poor prognosis set based on all preset dependent factor sets for each poor prognosis in the poor prognosis set, including: Based on each preset dependent factor set of each poor prognosis in the poor prognosis set and the single correlation values between all influencing factors in each preset dependent factor set and poor prognosis, selecting the optimal subset of each preset dependent factor set of each poor prognosis in the poor prognosis set; The optimal subsets of all preset dependent factor sets for each poor prognosis in the poor prognosis set are not empty sets, and all optimal subsets are extracted. Based on all the extracted optimal subsets, the dependent prognostic factor set for each poor prognosis in the poor prognosis set is determined.

[0059] In this embodiment, based on each preset dependent factor set for each poor prognosis in the poor prognosis set and the single correlation values between all influencing factors and poor prognosis in each preset dependent factor set, the optimal subset of each preset dependent factor set for each poor prognosis in the poor prognosis set is selected. The calculation formula can be expressed as: ; in, The optimal subset of the preset non-independent factor set of the b-th cluster set representing the a-th poor prognosis, represents the subset of the bth cluster set with the ath poor prognosis The multiple correlation values of It represents the subset of the first t influencing factors of the preset non-independent factor set of the b-th cluster set with the a-th poor prognosis, ST represents the preset multi-correlation threshold, and t represents the subset The number of influencing factors, represents the penalty coefficient based on the number of influencing factors in the subset, Representation subset The single correlation value between the cth influencing factor and the ath adverse prognosis, represents the cluster quality adjustment coefficient based on the average similarity of the bth cluster set with the ath poor prognosis, Represents the empty set.

[0060] In this embodiment, for each set of pre-set dependent factors for each adverse prognosis in the adverse prognosis set, the single correlation values (an indicator measuring the degree of linear correlation between a factor and adverse prognosis) of all influencing factors within the set are used for screening. Specifically, the factors are sorted from largest to smallest by the absolute value of the single correlation values, and a threshold is set (e.g., only factors with an absolute value of the single correlation value exceeding a certain value are retained) or a certain number of factors ranked high are selected. Factors with a weaker impact on adverse prognosis are eliminated, thereby obtaining an optimal subset of each pre-set dependent factor set, ensuring that the factors in the subset have a strong influence and correlation on adverse prognosis.

[0061] In this embodiment, the optimal subsets of all preset non-independent factor sets for each poor prognosis in the poor prognosis set are checked, and empty sets are eliminated (i.e., the correlation between all factors in the preset set and poor prognosis does not meet the standard).

[0062] In this embodiment, the remaining non-empty optimal subsets are integrated to form a set of dependent prognostic factors for each poor prognosis in the poor prognosis set. This set integrates the key dependent factors screened from multiple preset sets.

[0063] The beneficial effects of the above technology are: based on all preset non-independent factor sets for each poor prognosis in the poor prognosis set, the non-independent prognostic factor set for each poor prognosis in the poor prognosis set is determined, which can improve the scientificity and effectiveness of non-independent factor screening, more accurately reveal the synergistic effect between factors, provide a more reliable basis for postoperative risk warning, and enhance the accuracy of risk assessment and clinical application value. Example 7

[0064] Based on Example 1, a postoperative risk warning method based on machine learning is implemented based on historical patient data, a set of all independent prognostic factors for poor prognosis in a poor prognosis set, and a set of dependent prognostic factors to achieve postoperative risk warning, including: Based on the set of all independent prognostic factors and non-independent prognostic factors in the poor prognosis set, a postoperative risk classification and early warning model was constructed; The preoperative physiological data and intraoperative operation data of all historical patient sub-data in the historical patient data are used as the input of the postoperative risk classification warning model, and the postoperative physiological data of all historical patient sub-data in the historical patient data are used as the output of the postoperative risk classification warning model to train the postoperative risk classification warning model; Based on the trained postoperative risk classification and warning model, patients' postoperative risk warning can be achieved.

[0065] In this embodiment, the core elements are the set of independent prognostic factors and the set of dependent prognostic factors for all poor prognoses in the poor prognosis set. Using machine learning algorithms (such as logistic regression and random forest) or deep learning models (such as neural networks), a model framework is established, and the model structure and parameters are designed to enable the model to process data of different dimensions and analyze the comprehensive impact of factors on postoperative risk.

[0066] In this example, preoperative physiological data and intraoperative operation counts from all historical patient sub-data in the historical patient data are used as model input data. Postoperative physiological data are used as output labels. The model is iteratively trained using a large amount of historical data, with model parameters adjusted and performance optimized to accurately learn the mapping relationship between preoperative and intraoperative data and postoperative outcomes.

[0067] In this embodiment, the preoperative physiological data and intraoperative operation data of a new patient are input into the trained model. The model analyzes and processes the input data based on the learned rules to predict the patient's adverse prognosis after surgery.

[0068] The beneficial effects of the above technology are: based on historical patient data, the set of all independent prognostic factors for adverse prognosis in the adverse prognosis set, and the set of non-independent prognostic factors, postoperative risk warning can be achieved, which can comprehensively cover the dimensions of postoperative risk impact, accurately capture the preoperative-intraoperative-postoperative data correlation, and realize the intelligent transformation from data to risk prediction, providing clinicians with more accurate and comprehensive postoperative risk warnings, and significantly improving the scientificity and effectiveness of risk prevention and control. Example 8

[0069] The present invention provides a postoperative risk warning system based on machine learning, which is used to implement any one of the postoperative risk warning methods based on machine learning in Examples 1 to 7, with reference to Figure 2 ,include: Acquisition module: acquiring historical patient data of multiple patients who have undergone surgeries to be warned, analyzing the historical patient data to determine a poor prognosis set for the surgeries to be warned, and poor data for each poor prognosis in the poor prognosis set; Independent prognosis module: Based on historical patient data, poor prognosis set and poor data of each poor prognosis in the poor prognosis set, determine the influencing factor vector of each poor prognosis in the poor prognosis set and the independent prognostic factor set; Preset non-independent module: based on the influencing factor vector of each poor prognosis in the poor prognosis set and the independent prognosis factor set, determining multiple preset non-independent factor sets for each poor prognosis in the poor prognosis set; Dependent prognosis module: determining a dependent prognostic factor set for each poor prognosis in the poor prognosis set based on all preset dependent factor sets for each poor prognosis in the poor prognosis set; Early warning module: Based on historical patient data, the set of all independent prognostic factors for adverse prognosis in the adverse prognosis set, and the set of non-independent prognostic factors, postoperative risk warning is achieved.

[0070] The beneficial effects of the above technology are: by acquiring and analyzing historical patient data, determining the adverse prognosis set of surgeries to be warned, as well as the adverse data of each adverse prognosis in the adverse prognosis set, determining the influencing factor vector and independent prognostic factor set of each adverse prognosis in the adverse prognosis set, determining multiple preset non-independent factor sets for each adverse prognosis in the adverse prognosis set, determining the non-independent prognostic factor set for each adverse prognosis in the adverse prognosis set, and realizing postoperative risk warning based on historical patient data, the independent prognostic factor set of all adverse prognoses in the adverse prognosis set, and the non-independent prognostic factor set. It can comprehensively cover the dimensions of postoperative risk impact, accurately capture the preoperative, intraoperative, and postoperative data associations, realize the intelligent transformation from data to risk prediction, provide more accurate and comprehensive postoperative risk warnings for clinicians, and improve the scientificity, accuracy, and reliability of postoperative risk warnings.

[0071] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A postoperative risk warning method based on machine learning, characterized in that: include: S1: Obtain historical patient data of multiple patients who have undergone surgeries to be warned, analyze the historical patient data to determine a poor prognosis set for the surgeries to be warned, and poor data for each poor prognosis in the poor prognosis set; S2: Based on historical patient data, the poor prognosis set, and the poor data of each poor prognosis in the poor prognosis set, determine the influencing factor vector of each poor prognosis in the poor prognosis set and the independent prognostic factor set; S3: determining a plurality of preset non-independent factor sets for each poor prognosis in the poor prognosis set based on the influencing factor vector of each poor prognosis in the poor prognosis set and the independent prognosis factor set; S4: determining a set of independent prognostic factors for each poor prognosis in the poor prognosis set based on all preset independent factor sets for each poor prognosis in the poor prognosis set; S5: Achieve postoperative risk warning based on historical patient data, the set of all independent prognostic factors for adverse prognosis in the adverse prognosis set, and the set of non-independent prognostic factors.

2. A postoperative risk warning method based on machine learning according to claim 1, characterized in that: Access historical patient data for multiple patients who underwent surgeries requiring alerts, including: Obtain historical patient sub-data of each patient who underwent a surgery to be warned within a specified time period, wherein the historical patient sub-data includes patient information data, preoperative physiological data, intraoperative operation data, and postoperative physiological data; The historical patient data is determined based on the historical patient sub-data of all patients who underwent the surgery to be warned within a specified time period.

3. A postoperative risk warning method based on machine learning according to claim 2, characterized in that: Analyze historical patient data to determine a set of adverse prognoses for surgeries to be alerted, as well as adverse data for each adverse prognosis in the adverse prognosis set, including: Determine whether there is an adverse event in the postoperative physiological data of the historical patient sub-data of each patient in the historical patient data; if so, extract the adverse data from the postoperative physiological data of the historical patient sub-data of each patient in the historical patient data, wherein the adverse data includes at least one or more adverse sub-data, and the adverse sub-data includes the event patient, the adverse event, the adverse event occurrence time, and the adverse event severity; The extracted adverse data of all patients are statistically analyzed to determine the adverse prognosis set of the surgery to be warned, and the adverse data of each adverse prognosis in the adverse prognosis set are determined, wherein the adverse data includes multiple adverse sub-data of the corresponding adverse events.

4. A postoperative risk warning method based on machine learning according to claim 3, characterized in that: Based on historical patient data, a poor prognosis set, and poor data of each poor prognosis in the poor prognosis set, an influencing factor vector of each poor prognosis in the poor prognosis set and an independent prognostic factor set are determined, including: Performing feature extraction on each bad sub-data in the bad data of each bad prognosis in the bad prognosis set, and determining a bad feature vector of each bad sub-data in the bad data of each bad prognosis in the bad prognosis set; Based on the event patient in each adverse sub-data in each adverse data of each adverse prognosis in the adverse prognosis set, feature extraction is performed on the patient information data, preoperative physiological data, and intraoperative operation data of the corresponding historical patient sub-data in the historical patient data to determine the influencing factor vector of each adverse prognosis in the adverse prognosis set and the influencing factor value vector of each adverse sub-data in each adverse data of the adverse prognosis in the adverse prognosis set; Determine the first independent prognostic factor subset of each poor prognosis in the poor prognosis set based on the poor feature vectors, influencing factor value vectors, and Cox proportional hazard model of all poor sub-data in the poor prognosis data of each poor prognosis in the poor prognosis set, and record the number of first influencing factors in the first independent prognostic factor subset; Performing Pearson correlation analysis on the bad feature vectors and influencing factor value vectors of all bad sub-data in each bad data of bad prognosis in the bad prognosis set, and determining the single correlation value between each influencing factor in the influencing factor vector and the bad prognosis; Sorting the single correlation values of all influencing factors and poor prognosis in each poor prognosis influencing factor vector in the poor prognosis set from large to small, and selecting the first number of influencing factors after sorting to determine the second independent prognostic factor subset of each poor prognosis in the poor prognosis set; An independent prognostic factor set for each poor prognosis in the poor prognosis set is determined based on the first independent prognostic factor subset and the second independent prognostic factor subset for each poor prognosis in the poor prognosis set, wherein the independent prognostic factor set includes multiple influencing factors.

5. The postoperative risk warning method based on machine learning according to claim 4, characterized in that: Based on the influencing factor vector of each poor prognosis in the poor prognosis set and the independent prognosis factor set, multiple preset non-independent factor sets for each poor prognosis in the poor prognosis set are determined, including: Eliminating all influencing factors in the independent prognostic factor set from the influencing factor vector of each poor prognosis in the poor prognosis set, and determining a candidate set of non-independent factors for each poor prognosis in the poor prognosis set, wherein the candidate set of non-independent factors includes multiple influencing factors other than the independent prognostic factor set; Performing cluster analysis on all influencing factors in each candidate set of non-independent factors for poor prognosis in the poor prognosis set, determining multiple cluster sets based on the cluster analysis results, and recording the cluster similarity and the number of second influencing factors of each cluster set; Calculating the number of preset factors for each cluster set based on the cluster similarity of each cluster set of each poor prognosis in the poor prognosis set and the number of second influencing factors; All influencing factors in each cluster set of each poor prognosis in the poor prognosis set are sorted from large to small based on the single correlation value with poor prognosis, and the first preset number of influencing factors after sorting of each cluster set are selected as the preset non-independent factor set of each cluster set; Based on the preset dependent factor sets of all cluster sets for each poor prognosis in the poor prognosis set, multiple preset dependent factor sets for each poor prognosis in the poor prognosis set are determined.

6. A postoperative risk warning method based on machine learning according to claim 4, characterized in that: Based on all preset dependent factor sets for each poor prognosis in the poor prognosis set, a dependent prognostic factor set for each poor prognosis in the poor prognosis set is determined, including: Based on each preset dependent factor set of each poor prognosis in the poor prognosis set and the single correlation values between all influencing factors in each preset dependent factor set and poor prognosis, selecting the optimal subset of each preset dependent factor set of each poor prognosis in the poor prognosis set; The optimal subsets of all preset dependent factor sets for each poor prognosis in the poor prognosis set are not empty sets, and all optimal subsets are extracted. Based on all the extracted optimal subsets, the dependent prognostic factor set for each poor prognosis in the poor prognosis set is determined.

7. The postoperative risk warning method based on machine learning according to claim 1, characterized in that: Based on historical patient data, all independent prognostic factor sets for adverse prognosis in the adverse prognosis set, and a set of non-independent prognostic factors, postoperative risk warning is achieved, including: Based on the set of all independent prognostic factors and non-independent prognostic factors in the poor prognosis set, a postoperative risk classification and early warning model was constructed; The preoperative physiological data and intraoperative operation data of all historical patient sub-data in the historical patient data are used as the input of the postoperative risk classification warning model, and the postoperative physiological data of all historical patient sub-data in the historical patient data are used as the output of the postoperative risk classification warning model to train the postoperative risk classification warning model; Based on the trained postoperative risk classification and warning model, patients' postoperative risk warning can be achieved.

8. A postoperative risk warning system based on machine learning, characterized in that: A method for performing a postoperative risk warning method based on machine learning according to any one of claims 1 to 7, comprising: Acquisition module: acquiring historical patient data of multiple patients who have undergone surgeries to be warned, analyzing the historical patient data to determine a poor prognosis set for the surgeries to be warned, and poor data for each poor prognosis in the poor prognosis set; Independent prognosis module: Based on historical patient data, poor prognosis set and poor data of each poor prognosis in the poor prognosis set, determine the influencing factor vector of each poor prognosis in the poor prognosis set and the independent prognostic factor set; Preset non-independent module: based on the influencing factor vector of each poor prognosis in the poor prognosis set and the independent prognosis factor set, determining multiple preset non-independent factor sets for each poor prognosis in the poor prognosis set; Dependent prognosis module: determining a dependent prognostic factor set for each poor prognosis in the poor prognosis set based on all preset dependent factor sets for each poor prognosis in the poor prognosis set; Early warning module: Based on historical patient data, the set of all independent prognostic factors for adverse prognosis in the adverse prognosis set, and the set of non-independent prognostic factors, postoperative risk warning is achieved.

Images