Prediction method for intraoperative hypertension

Abstract

The invention relates to a method for predicting intraoperative hypertension, and belongs to the field of digital medical treatment. Comprising the following steps: preprocessing, standardizing and marking collected patient data to obtain first data; selecting characteristic variables from the first data, and selecting influence characteristics from the characteristic variables according to correlation coefficient calculation and expert judgment; establishing a sample set based on the influence characteristics, and training through the sample set to obtain a trained intraoperative hypertension prediction model; preprocessing and standardizing data of a patient to be tested to obtain second data, and inputting the second data into the trained intraoperative hypertension prediction model to realize prediction of intraoperative hypertension of the patient. According to the method, the influence factors of intraoperative hypertension are searched from a large amount of data, and the intraoperative hypertension prediction model is trained based on the log-likelihood function, so that the accuracy of intraoperative hypertension prediction is improved, and the problem that current intraoperative hypertension prediction depends on clinical experience of doctors is solved.

Description

Technical Field

[0001] The present invention relates to the field of digital medical technology, and in particular to a method for predicting intraoperative hypertension. Background Art

[0002] Intraoperative hypertension is a common clinical disease that has a significant impact on the patient's health status and surgical outcomes. Traditional methods for predicting intraoperative hypertension rely on the physician's clinical experience and are highly subjective, resulting in low efficiency and inaccurate prediction results. In addition, data related to intraoperative hypertension prediction usually come from different medical systems and have different recording formats, which makes data integration and analysis difficult, further reducing work efficiency.

[0003] Therefore, it is of great practical significance to provide an automated method to discover, clean and standardize data related to intraoperative hypertension and predict intraoperative hypertension based on these data. Summary of the invention

[0004] In view of the above analysis, the present invention aims to provide a method for predicting intraoperative hypertension, so as to solve the problem that the current prediction of intraoperative hypertension relies on the clinical experience of doctors, which is inefficient and the prediction results are not accurate enough.

[0005] The present invention provides a method for predicting intraoperative hypertension, the method comprising the following steps:

[0006] Collect historical medical record information of multiple patients through an electronic medical record system, collect historical physiological parameter monitoring data of the multiple patients through medical equipment or scanning monitoring results, and pre-process the historical medical record information and historical physiological parameter monitoring data of the multiple patients; build a rule base, and standardize and annotate the pre-processed data through the rule base to obtain first data;

[0007] Selecting a field related to intraoperative hypertension from the first data as a characteristic variable, selecting a "preoperative blood pressure" field as a target variable, and selecting an influencing feature from the characteristic variables based on correlation coefficient calculation and expert judgment;

[0008] Constructing an intraoperative hypertension prediction model, selecting the influencing feature and data values ​​corresponding to the target variable from the first data to establish a sample set, and training the intraoperative hypertension prediction model through the sample set to obtain a trained intraoperative hypertension prediction model;

[0009] According to the influencing characteristics of intraoperative hypertension and the target variables, the historical medical information of the patient to be tested and the corresponding data in the historical physiological parameter monitoring data are collected, and preprocessed and standardized to obtain the second data; the second data is input into the trained intraoperative hypertension prediction model to realize the prediction of the patient's intraoperative hypertension.

[0010] 3. Further, the preprocessing of the historical medical record information and historical physiological parameter monitoring data of the multiple patients includes:

[0011] The historical medical record information and historical physiological parameter monitoring data of the multiple patients are stored as a data table according to the patients and the time of consultation. The data table includes fields, data values ​​corresponding to the fields, and value ranges of the fields; wherein the data of the same patient visiting the doctor on the same day is one record in the data table;

[0012] When the data value corresponding to a field exceeds the value range of the field, the record corresponding to the data value is deleted to remove the abnormal value;

[0013] By calculating the first similarity between the fields, the fields whose first similarity is greater than the corresponding threshold are merged into one field; by calculating the data fingerprint of each record, for each record, based on the data fingerprint of each record, the second similarity between the record and other records is calculated, and one record is selected from the records whose second similarity is greater than the corresponding threshold and the other records are deleted to remove redundant data;

[0014] When the data value corresponding to a field is missing, the data value is automatically inferred or supplemented manually.

[0015] Further, the step of calculating the data fingerprint of each record, calculating the second similarity between the record and other records based on the data fingerprint of each record, selecting a record from the records whose second similarity is greater than a corresponding threshold and deleting other records to remove redundant data includes:

[0016] Step S11, setting the weight of each field;

[0017] Step S12: For each record, obtain the Hash value of the data value corresponding to each field of the record through the MD5 function, and obtain the data fingerprint of the record based on the Hash value of the data value corresponding to each field and its weight;

[0018] Step S13: For each record, calculate the Hamming distance between the data fingerprint of the record and the data fingerprints of other records to obtain a second similarity, and select the record with the least number of missing data values ​​corresponding to the field from the records whose second similarity is greater than the corresponding threshold; if there are multiple records with the least number of missing data values ​​corresponding to the fields, select the record with the most recent time and delete the other records.

[0019] Further, after encoding the Hash value of the data value corresponding to each field, each bit of the Hash value is multiplied by its corresponding weight to obtain a weighted Hash value;

[0020] The weighted Hash value of each field is accumulated according to the corresponding bit to obtain the SimHash value of the record;

[0021] The SimHash value is converted into binary to obtain the data fingerprint of the record.

[0022] Furthermore, the second similarity calculation is performed by the following method:

[0023] R=1-(H(K1,K2) / W),

[0024] Among them, R is the similarity, K1 and K2 are two records, H(K1, K2) means calculating the Hamming distance between K1 and K2, and W is the number of bits of the data fingerprint.

[0025] Furthermore, the construction of the rule base includes:

[0026] For the data value corresponding to each field, a mapping rule is defined based on a regular expression, and the data value is mapped into a standardized expression based on the mapping rule;

[0027] Building a list based on the standardized expression and corresponding fields;

[0028] The mapping rules and the list are stored in a database as a rule base.

[0029] Furthermore, the selecting of influencing features from the feature variables according to correlation coefficient calculation and expert judgment includes:

[0030] The characteristic variable is divided into a first characteristic variable and a second characteristic variable, and the correlation coefficient between the data value corresponding to the first characteristic variable and the data value corresponding to the target variable is calculated, and the first characteristic variable corresponding to the correlation coefficient greater than the threshold is selected as the first influencing feature; and the relevant characteristic variable is selected from the second characteristic variable as the second influencing feature according to the expert judgment;

[0031] The first characteristic variables include age, BMI index, smoking, alcoholism, sodium intake, blood sugar, blood lipids, history of hypertension, history of diabetes, family history of hypertension, and heart rate; the second characteristic variables include use of pressor drugs, use of hormonal drugs, use of non-steroidal anti-inflammatory drugs, use of oral contraceptives, use of antidepressants, and use of anti-tumor drugs.

[0032] Furthermore, the correlation coefficient between the data value corresponding to the first characteristic variable and the data value corresponding to the target variable is calculated by the following formula:

[0033]

[0034] Among them, λ iis the correlation coefficient between the data value corresponding to the i-th first feature variable and the data value corresponding to the target variable, F i is the data value corresponding to the first characteristic variable of the i-th order, T is the data value corresponding to the target variable, cov(F i ,T) is F i Covariance with T, σF i F i is the standard deviation of , σT is the standard deviation of T.

[0035] Furthermore, the intraoperative hypertension prediction model includes an input layer, a linear combination layer and an output layer; the input layer is used to receive the characteristic values ​​of the samples; the linear combination layer includes a linear combination unit and a mapping function, the linear combination unit is used to perform weighted summation on the input characteristic values, and the mapping function is used to map the output of the linear combination unit to the interval (0,1); the output layer is used to output the predicted probability of the patient developing intraoperative hypertension and the corresponding prediction label.

[0036] Furthermore, the predicted probability of intraoperative hypertension in patients was obtained by the following formula:

[0037]

[0038] Among them, P i represents the predicted probability of intraoperative hypertension in the ith patient; exp represents the exponential function; W ij is the weight of the jth influencing feature of the i-th patient; X ij is the data value corresponding to the jth influencing feature of the ith patient; n is the number of influencing features, that is, the sum of the first influencing feature, the second influencing feature and the target variable; W0 is the bias.

[0039] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0040] 1. The present invention collects information of surgical patients from multiple medical institutions, then searches for influencing factors of intraoperative hypertension from a large amount of data and trains an intraoperative hypertension prediction model based on the log-likelihood function, thereby improving the accuracy of intraoperative hypertension prediction.

[0041] 2. The present invention eliminates the problems of data redundancy, anomaly and missing by preprocessing the data; standardizes the preprocessed data by constructing a rule base, thereby unifying the terminology of data from different medical institutions, eliminating the problem of inconsistent terminology, improving the quality and availability of data, and improving the efficiency of data integration and analysis, providing a reliable foundation for the subsequent training and verification of the intraoperative hypertension model.

[0042] 3. The present invention automatically predicts and marks patients who will develop intraoperative hypertension through a trained intraoperative hypertension prediction model, thereby achieving early detection and prevention of intraoperative hypertension and improving work efficiency and treatment effects.

[0043] In the present invention, the above-mentioned technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent description, and some advantages can become obvious from the description, or can be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like components throughout the drawings.

[0045] Figure 1 This is a flow chart of a method for predicting intraoperative hypertension according to an embodiment of the present invention;

[0046] Figure 2 Schematic diagram of the structure of the intraoperative hypertension prediction model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0048] A specific embodiment of the present invention discloses a method for predicting intraoperative hypertension. Figure 1 As shown, the method comprises the following steps:

[0049] Step S1, collecting historical medical record information of multiple patients through an electronic medical record system, collecting historical physiological parameter monitoring data of the multiple patients through medical equipment or scanning monitoring results, and preprocessing the historical medical record information and historical physiological parameter monitoring data of the multiple patients; building a rule base, and standardizing and annotating the preprocessed data through the rule base to obtain first data;

[0050] Step S2, selecting fields related to intraoperative hypertension from the first data as feature variables, selecting the "preoperative blood pressure" field as the target variable, and selecting influencing features from the feature variables based on correlation coefficient calculation and expert judgment;

[0051] Step S3, constructing an intraoperative hypertension prediction model, selecting the influencing features and data values ​​corresponding to the target variable from the first data to establish a sample set, and training the intraoperative hypertension prediction model through the sample set to obtain a trained intraoperative hypertension prediction model;

[0052] Step S4, according to the influencing characteristics of the intraoperative hypertension and the target variables, collect the historical medical information of the patient to be tested and the corresponding data in the historical physiological parameter monitoring data, and preprocess and standardize them to obtain the second data; input the second data into the trained intraoperative hypertension prediction model to realize the prediction of the intraoperative hypertension of the patient.

[0053] Specifically, in step S1, the medical institution's electronic medical record system stores the historical medical record information of the patients, which includes the patient's personal information, medical information, examination / imaging information, medical advice information, diagnosis information, treatment information, medical history information, hospitalization records, and subsequent follow-up information.

[0054] Specifically, part of the patient's historical physiological parameter monitoring data is directly extracted from the test result storage system (such as electrocardiogram, blood pressure and other data), and the other part is collected by scanning paper monitoring results using OCR technology.

[0055] Specifically, the patient's medical history information refers to diseases related to intraoperative hypertension, including: cardiovascular system diseases, endocrine system diseases, kidney diseases, nervous system diseases, respiratory system diseases, and other system diseases; among them, cardiovascular system diseases include hypertension, heart disease, and heart failure, endocrine system diseases include diabetes and thyroid diseases, kidney diseases include chronic kidney disease and renal artery stenosis, nervous system diseases include cerebrovascular diseases and autonomic dysfunction, respiratory system diseases include chronic obstructive pulmonary disease and sleep apnea syndrome, and other system diseases include rheumatic immune diseases and blood system diseases. The patient's examination / imaging information includes renal function (serum creatinine, urea nitrogen, uric acid, glomerular filtration rate), blood lipids (total cholesterol, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol), blood sugar (fasting blood sugar, 2-hour postprandial blood sugar, glycosylated hemoglobin), electrolytes (sodium, potassium, chloride, calcium, phosphorus), inflammatory indicators (C-reactive protein, white blood cell count), and others (renin, angiotensin 2, aldosterone). The patient's physiological parameter monitoring data includes electrocardiogram, heart rate, blood pressure, blood oxygen saturation, body temperature, and respiratory rate. The physiological parameter monitoring data includes not only the monitoring data during the operation, but also the monitoring data obtained through examination.

[0056] It should be noted that intraoperative hypertension refers to hypertension that occurs in patients during surgery. Intraoperative hypertension is different from ordinary hypertension and is usually temporary. It may be caused by various factors such as surgical stimulation and the patient's medical history. Since intraoperative hypertension can easily lead to surgical complications, it is necessary to predict the probability of intraoperative hypertension in patients and take preventive measures to ensure the safety of the operation. It is understandable that the patient's personal information (such as age, BMI index, smoking, alcoholism, and sodium intake) will affect the probability of intraoperative hypertension. The patient's medical information (condition), examination / imaging information, and physiological parameter monitoring data directly determine the doctor's medical advice, diagnosis, and treatment information. The patient's medical history information, hospitalization records, and subsequent follow-up information determine the patient's basic physical condition.

[0057] Furthermore, the preprocessing of the historical medical record information and historical physiological parameter monitoring data of the multiple patients includes:

[0058] The historical medical record information and historical physiological parameter monitoring data of the multiple patients are stored as a data table according to the patients and the time of consultation. The data table includes fields, data values ​​corresponding to the fields, and value ranges of the fields; wherein the data of the same patient visiting the doctor on the same day is one record in the data table;

[0059] When the data value corresponding to a field exceeds the value range of the field, the record corresponding to the data value is deleted to remove the abnormal value;

[0060] By calculating the first similarity between the fields, the fields whose first similarity is greater than the corresponding threshold are merged into one field; by calculating the data fingerprint of each record, for each record, based on the data fingerprint of each record, the second similarity between the record and other records is calculated, and one record is selected from the records whose second similarity is greater than the corresponding threshold and the other records are deleted to remove redundant data;

[0061] When the data value corresponding to a field is missing, the data value is automatically inferred or supplemented manually.

[0062] Specifically, for each field in the data table, its value range is determined according to its meaning and actual situation. When the data value corresponding to a field exceeds the value range of the field, the record corresponding to the data value is deleted. The cosine similarity between the fields is calculated as the first similarity. When the fields with the first similarity greater than the corresponding threshold are merged into one field, if the merged fields have corresponding data values ​​in one record, the data with the most recent time is used as the data value corresponding to the merged field.

[0063] Further, the step of calculating the data fingerprint of each record, calculating the second similarity between the record and other records based on the data fingerprint of each record, selecting a record from the records whose second similarity is greater than a corresponding threshold and deleting other records to remove redundant data includes:

[0064] Step S11, setting the weight of each field;

[0065] Step S12: For each record, obtain the Hash value of the data value corresponding to each field of the record through the MD5 function, and obtain the data fingerprint of the record based on the Hash value of the data value corresponding to each field and its weight;

[0066] Step S13: For each record, calculate the Hamming distance between the data fingerprint of the record and the data fingerprints of other records to obtain a second similarity, and select the record with the least number of missing data values ​​corresponding to the field from the records whose second similarity is greater than the corresponding threshold; if there are multiple records with the least number of missing data values ​​corresponding to the fields, select the record with the most recent time and delete the other records.

[0067] Specifically, in step S11, the weight of each field is set according to business experience, and the greater the weight, the more important the field is. The weight is a positive integer.

[0068] Specifically, in step S12, after the Hash value of the data value corresponding to each field is encoded (for example, 0 is encoded as -1), each bit of the Hash value is multiplied by its corresponding weight to obtain a weighted Hash value;

[0069] The weighted Hash value of each field is accumulated according to the corresponding bit to obtain the SimHash value of the record;

[0070] The SimHash value is converted into binary to obtain the data fingerprint of the record.

[0071] Specifically, in step S13, the second similarity calculation is performed by the following method:

[0072] R=1-(H(K1,K2) / W),

[0073] Among them, R is the similarity, K1 and K2 are two records, H(K1, K2) means calculating the Hamming distance between K1 and K2, and W is the number of bits of the data fingerprint.

[0074] It can be understood that the present application measures the similarity of two recorded data fingerprints by calculating their Hamming distance. The smaller the Hamming distance, the higher the similarity.

[0075] Specifically, when the data value corresponding to a field is missing, if the data value corresponding to the field can be inferred from the data values ​​corresponding to other fields, it is automatically inferred and supplemented according to the inference result; otherwise, the data value is supplemented manually.

[0076] For example, assume that the collected original data of the patient is as shown in Table 1, wherein the data value corresponding to the "sex" field of the patient in the records with IDs 1 and 3 is missing. The data value corresponding to the "sex" field can be inferred from the data value corresponding to the "menstrual history" field. The data value corresponding to the "menstrual history" field is 1, which represents menstrual history. Therefore, the data value corresponding to the "sex" field of the patient in the records with IDs 1 and 3 is automatically inferred to be "female", and the data value is supplemented with "female".

[0077] Table 1 Original data of patients collected

[0078] ID gender … Menstrual history Marital status Pregnancy history 1 / … 1 not yet 1 2 male … 0 not yet 1 3 / … 1 already 0

[0079] Furthermore, the construction of the rule base includes:

[0080] For the data value corresponding to each field, a mapping rule is defined based on a regular expression, and the data value is mapped into a standardized expression based on the mapping rule;

[0081] Building a list based on the standardized expression and corresponding fields;

[0082] The mapping rules and the list are stored in a database as a rule base.

[0083] It should be noted that the data value corresponding to each field is a term in the medical scenario. This application defines specific mapping rules for each of the above terms to map them into standardized expressions, where the disease name is mapped to the corresponding code according to the International Classification of Diseases coding table, the diagnosis result is mapped to the corresponding code according to the standard diagnosis coding table, and the type of surgery is mapped to the corresponding code according to the International Classification of Surgery coding table. These standardized expressions are verified, and a standardized term list is constructed based on the verified standardized expressions and the corresponding fields, and the mapping rules and standardized term list are stored in the database as a rule base.

[0084] For example, as shown in Table 2, mapping rules are defined for the data values ​​corresponding to "sex", "disease name", "drug name", "diagnosis result" and "surgery type" based on regular expressions. Based on these mapping rules, sex is mapped to "M" or "F", stage 2 hypertension is mapped to "ICD-10:I10" according to the International Classification of Diseases coding table, aspirin is mapped to "aspirin", the diagnosis result of coronary artery atherosclerotic heart disease is mapped to "ICD-10:125.1" according to the standard diagnosis coding table, and appendectomy is mapped to "ICD-10-PCS:0DTJ4ZZ" according to the International Classification of Surgery coding table. A standardized term list is constructed based on the above standardized expressions and corresponding fields, as shown in Table 3.

[0085] Table 2 Mapping rules

[0086]

[0087]

[0088] Table 3 List of standardized terms

[0089] Fields Standardized expression illustrate gender M Indicates male gender F Indicates female Disease name ICD-10:I10 Indicates (essential) hypertension Drug Name aspirin English abbreviation: ASA Diagnosis ICD-10:125.1 Coronary atherosclerotic heart disease Type of surgery ICD-10-PCS:0DTJ4ZZ Appendectomy

[0090] It can be understood that the data value corresponding to the "Disease Name" field is a variety of specific diseases, and corresponding mapping rules need to be defined based on the terms of each disease. The above example only shows the mapping rules for common primary hypertension diseases.

[0091] Furthermore, the step of standardizing and labeling the preprocessed data through the rule base to obtain the first data includes:

[0092] Mapping the preprocessed data into a standardized expression through the rule base;

[0093] The mapped data is standardized in terms of date, time, value and unit according to the predetermined data format specification;

[0094] The data with unified format, that is, each record in the data table, is marked according to whether intraoperative hypertension occurs to obtain the first data.

[0095] Specifically, the predetermined data format specification includes date format, time format, decimal point digits, unified international units, etc. The label for intraoperative hypertension is 1, otherwise the label is 0.

[0096] It should be noted that the measurement units in different data sources need to be converted into unified international units, and unit conversions are performed as needed, such as converting blood pressure values ​​from mmHg to kPa, converting blood sugar values ​​from mg / dL to mmol / L, etc. Since the patient information collected from a single medical institution has limitations, this application collects information on surgical patients from multiple medical institutions, and then finds the influencing factors of intraoperative hypertension from a large amount of data and trains an intraoperative hypertension prediction model based on a large amount of data to improve the accuracy of intraoperative hypertension prediction. Since the collected patient information comes from different medical institutions and different systems of the same medical institution, and the term definitions and field definitions of each medical institution are different, the collected raw data not only has inconsistent terms, but also has redundancy, anomalies, and missing problems. This application eliminates the problems of data redundancy, anomalies, and missing by preprocessing the data; standardizes the preprocessed data by building a rule base, thereby unifying the terms of the data from different medical institutions, eliminating the problem of inconsistent terms, and improving the quality and availability of the data.

[0097] Specifically, in step S2, the selecting of influencing features from the feature variables according to correlation coefficient calculation and expert judgment includes:

[0098] The characteristic variable is divided into a first characteristic variable and a second characteristic variable, the correlation coefficient between the data value corresponding to the first characteristic variable and the data value corresponding to the target variable is calculated, and the first characteristic variable corresponding to the correlation coefficient greater than the threshold is selected as the first influencing feature; and the relevant characteristic variable is selected from the second characteristic variable as the second influencing feature according to expert judgment.

[0099] Specifically, the first characteristic variables include age, BMI index, smoking, alcoholism, sodium intake, blood sugar, blood lipids, history of hypertension, history of diabetes, family history of hypertension, and heart rate; the second characteristic variables include whether pressor drugs are used, whether hormonal drugs are used, whether non-steroidal anti-inflammatory drugs are used, whether oral contraceptives are used, whether antidepressants are used, and whether anti-tumor drugs are used.

[0100] Specifically, pressor drugs include dopamine, norepinephrine, and epinephrine; hormonal drugs include glucocorticoids such as dexamethasone, prednisone, and methylprednisolone; non-steroidal anti-inflammatory drugs include ibuprofen and naproxen; antidepressants include monoamine oxidase inhibitors and tricyclic antidepressants; and anti-tumor drugs include paclitaxel and cisplatin.

[0101] Furthermore, the correlation coefficient between the data value corresponding to the first characteristic variable and the data value corresponding to the target variable is calculated by the following formula:

[0102]

[0103] Among them, λ i is the correlation coefficient between the data value corresponding to the i-th first feature variable and the data value corresponding to the target variable, F i is the data value corresponding to the first characteristic variable of the i-th order, T is the data value corresponding to the target variable, cov(F i ,T) is F i Covariance with T, σF i F i is the standard deviation of , σT is the standard deviation of T.

[0104] It should be noted that each record in the first data is the corresponding data of a patient. It is necessary to select a certain number of data values ​​corresponding to the first characteristic variable from the first data according to the actual situation to calculate the above correlation coefficient. The larger the correlation coefficient, the greater the correlation between the first characteristic variable and intraoperative hypertension. The first influencing feature obtained by the present invention through the above method based on the collected data is age, BMI index, smoking, alcoholism, sodium intake, blood sugar, blood lipids, history of hypertension, and history of diabetes. The second influencing feature obtained by the present invention based on expert judgment is whether to use pressor drugs, whether to use hormone drugs, whether to use non-steroidal anti-inflammatory drugs, and whether to use oral contraceptives.

[0105] Specifically, in step S3, the first influencing feature, the second influencing feature and the data values ​​corresponding to the target variable are selected from the first data to establish a sample set. Figure 2 As shown, the intraoperative hypertension prediction model includes an input layer, a linear combination layer and an output layer; the input layer is used to receive the characteristic value of the sample; the linear combination layer includes a linear combination unit and a mapping function, the linear combination unit is used to perform weighted summation on the input characteristic value, and the mapping function is used to map the output of the linear combination unit to the interval (0,1); the output layer is used to output the predicted probability of the patient developing intraoperative hypertension and the corresponding prediction label.

[0106] Furthermore, the predicted probability of intraoperative hypertension in patients was obtained by the following formula:

[0107]

[0108] Among them, P i represents the predicted probability of intraoperative hypertension in the ith patient; exp represents the exponential function; W ij is the weight of the jth influencing feature of the i-th patient; X ij is the data value corresponding to the jth influencing feature of the ith patient; n is the number of influencing features, that is, the sum of the first influencing feature, the second influencing feature and the target variable; W0 is the bias.

[0109] It should be noted that when the predicted probability of a patient developing intraoperative hypertension is greater than a preset threshold (eg, 0.5), the prediction label corresponding to the patient is set to 1, otherwise it is set to 0.

[0110] Furthermore, the expression of the i-th sample is: D i =(X i1 ,X i2 ,......X in ,Y i ), where X in represents the data value corresponding to the nth influencing feature of the i-th sample, where n is the number of influencing features, i.e., the sum of the first influencing feature, the second influencing feature, and the target variable; Y i Represents the annotation label of the i-th sample.

[0111] Furthermore, the step of training the intraoperative hypertension prediction model by using the sample set to obtain a trained intraoperative hypertension prediction model comprises:

[0112] Step S31, dividing the sample set into a training set and a test set;

[0113] Step S32: establishing a log-likelihood function based on the training set, and calculating the weight and bias of each influencing feature when the log-likelihood function is maximum;

[0114] Step S33, setting the calculation gradient of the weight and bias of each influencing feature, iteratively updating the weight and bias of each influencing feature through the calculation gradient, until the change of the weight and bias is less than a threshold or the maximum number of iterations is reached, and the iteration is stopped to obtain a trained intraoperative hypertension prediction model;

[0115] Step S34, input the test set into the trained intraoperative hypertension prediction model to obtain the predicted probability of intraoperative hypertension and the corresponding prediction label, calculate the accuracy and recall based on the corresponding prediction label and annotation label, if the accuracy and recall both reach the threshold, obtain the trained intraoperative hypertension prediction model; otherwise, expand the number of samples in the sample set or reduce the threshold corresponding to the predicted probability of intraoperative hypertension and return to step S31.

[0116] Specifically, in step S31, the sample set is divided into a training set and a test set at a ratio of 8:2.

[0117] Specifically, in step S32, the expression of the log-likelihood function is:

[0118]

[0119] Among them, l(W ij ,W0) represents the log-likelihood function, ln represents the logarithm with base e, Yi represents the annotation label of the i-th patient.

[0120] Find l(W ij ,W0) is the maximum weight W ij and W0, thereby obtaining the weight and bias of each influencing feature.

[0121] Specifically, in step S33, the calculation gradient of the weight of each influencing feature is set by the following formula:

[0122]

[0123] in, represents the predicted label of the ith patient, and m is the number of samples in the training set.

[0124] The calculation gradient of the bias is set by the following formula:

[0125]

[0126] Furthermore, the weight of each influencing feature is updated by the following formula:

[0127]

[0128] Among them, W ij (t+1) represents the updated weight of the jth influencing feature of the ith patient, W ij (t) represents the weight of the jth influencing feature of the current i-th patient, and α is a constant representing the learning rate.

[0129] Update the bias using the following formula:

[0130]

[0131] Among them, W0 (t+1) represents the updated bias of the ith patient, W0 (t) represents the bias of the current i-th patient.

[0132] Preferably, α is set to 0.01, and the change in weight and bias is set to be less than the threshold value 10 -6 Or the iteration stops when the maximum number of iterations, 1000, is reached.

[0133] It should be noted that when the iteration is stopped, the current weights and biases of the various influencing features are recorded, so as to obtain a trained intraoperative hypertension prediction model corresponding to the weights and biases.

[0134] Specifically, in step S34, the accuracy is calculated by the following formula:

[0135]

[0136] Among them, TP represents the number of correct predictions of intraoperative hypertension, and FP represents the number of incorrect predictions of intraoperative hypertension.

[0137] The recall rate is calculated by the following formula:

[0138]

[0139] Among them, TN represents the number of cases in which intraoperative hypertension was correctly predicted, and FN represents the number of cases in which intraoperative hypertension was not predicted but actually occurred.

[0140] It should be noted that the accuracy reflects the proportion of samples with correct predictions of intraoperative hypertension to the total number of samples, and the recall rate reflects the proportion of samples with correct predictions of intraoperative hypertension to all samples that actually experience intraoperative hypertension.

[0141] Furthermore, the threshold corresponding to expanding the number of samples in the sample set or reducing the predicted probability of intraoperative hypertension includes:

[0142] If the accuracy rate does not reach the threshold, the number of samples in the sample set is expanded, thereby expanding the number of samples in the training set; if the recall rate does not reach the threshold, the threshold corresponding to the predicted probability of intraoperative hypertension is lowered.

[0143] It can be understood that when the recall rate does not reach the threshold, it means that a large proportion of samples that actually experience intraoperative hypertension are not correctly predicted to have intraoperative hypertension. Therefore, the threshold corresponding to the predicted probability of intraoperative hypertension should be lowered so that the predicted labels of more samples are set to 1, that is, the number of samples predicted to have intraoperative hypertension is increased.

[0144] Specifically, in step S4, the historical medical record information of the patient to be tested and the corresponding data in the historical physiological parameter monitoring data are collected according to the first influencing feature, the second influencing feature and the target variable of the intraoperative hypertension described in step S2, and the second data are obtained by preprocessing and standardizing through the method in step S1; the second data is input into the intraoperative hypertension prediction model trained in step S3 to obtain the probability of intraoperative hypertension and the corresponding prediction label. When the prediction label of the patient to be tested is 1, he or she is marked as a patient at high risk of intraoperative hypertension in the medical management system.

[0145] Preferably, when the physician issues an application for surgery, the physician first predicts the patient's intraoperative hypertension through step S4. When the patient's prediction label is 1, that is, it is predicted that intraoperative hypertension will occur, the physician is given an early warning and the surgery application is controlled.

[0146] Specifically, the warning information includes the patient's age, BMI index, preoperative blood pressure, surgery type, medical history information, and medication information.

[0147] It should be noted that the present application marks patients who are predicted to develop intraoperative hypertension and issues early warning reminders. Even if the physician is subsequently changed, the physician will be warned for the patient, thereby improving the safety of medical practices.

[0148] Specifically, the control over surgical applications includes: requiring physicians to supplement relevant application materials and risk assessment materials before they can continue to submit surgical applications.

[0149] Preferably, when the predicted label of the patient is 1, a recommended diagnosis and treatment path for the patient is provided to the physician.

[0150] Specifically, the data values ​​corresponding to the patient's "diagnosis result", "surgery type", "affiliated department", and "gender" fields are retrieved in the database; based on the search results, similar patients whose data values ​​corresponding to the "diagnosis result", "surgery type", "affiliated department", and "gender" fields are the same as those of the patient and who are marked as at high risk of intraoperative hypertension are retrieved in the database; the medical order information of each similar patient is obtained based on its unique identifier (ID), and the medical order information of all similar patients is provided to the physician as the recommended diagnosis and treatment path for the patient.

[0151] It should be noted that medical order information includes examination, surgery, drug information, nursing measures, and dietary restrictions prescribed by the physician. The above-mentioned recommended treatment pathways assist physicians in providing targeted diagnosis and treatment, thereby improving the effectiveness of diagnosis and treatment.

[0152] Compared with the prior art, the method for predicting intraoperative hypertension and the system for early warning control and auxiliary diagnosis and treatment of intraoperative hypertension provided by the present invention have the following beneficial effects:

[0153] 1. The present invention collects information of surgical patients from multiple medical institutions, then searches for influencing factors of intraoperative hypertension from a large amount of data and trains an intraoperative hypertension prediction model based on the log-likelihood function, thereby improving the accuracy of intraoperative hypertension prediction.

[0154] 2. The present invention eliminates the problems of data redundancy, anomaly and missing by preprocessing the data; standardizes the preprocessed data by constructing a rule base, thereby unifying the terminology of data from different medical institutions, eliminating the problem of inconsistent terminology, improving the quality and availability of data, and improving the efficiency of data integration and analysis, providing a reliable foundation for the subsequent training and verification of the intraoperative hypertension model.

[0155] 3. The present invention automatically predicts and marks patients who will develop intraoperative hypertension through a trained intraoperative hypertension prediction model, thereby achieving early detection and prevention of intraoperative hypertension and improving work efficiency and treatment effects.

[0156] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0157] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for predicting intraoperative hypertension, characterized in that: The method comprises the following steps: Collect historical medical record information of multiple patients through an electronic medical record system, collect historical physiological parameter monitoring data of the multiple patients through medical equipment or scanning monitoring results, and pre-process the historical medical record information and historical physiological parameter monitoring data of the multiple patients; build a rule base, and standardize and annotate the pre-processed data through the rule base to obtain first data; Selecting a field related to intraoperative hypertension from the first data as a characteristic variable, selecting a "preoperative blood pressure" field as a target variable, and selecting an influencing feature from the characteristic variables based on correlation coefficient calculation and expert judgment; Constructing an intraoperative hypertension prediction model, selecting the influencing feature and data values ​​corresponding to the target variable from the first data to establish a sample set, and training the intraoperative hypertension prediction model through the sample set to obtain a trained intraoperative hypertension prediction model; According to the influencing characteristics of intraoperative hypertension and the target variables, the historical medical information of the patient to be tested and the corresponding data in the historical physiological parameter monitoring data are collected, and preprocessed and standardized to obtain the second data; the second data is input into the trained intraoperative hypertension prediction model to realize the prediction of the patient's intraoperative hypertension.

2. A method for predicting intraoperative hypertension according to claim 1, characterized in that: The preprocessing of the historical medical record information and historical physiological parameter monitoring data of the multiple patients includes: The historical medical record information and historical physiological parameter monitoring data of the multiple patients are stored as a data table according to the patients and the time of consultation. The data table includes fields, data values ​​corresponding to the fields, and value ranges of the fields; wherein the data of the same patient visiting the doctor on the same day is one record in the data table; When the data value corresponding to a field exceeds the value range of the field, the record corresponding to the data value is deleted to remove the abnormal value; By calculating the first similarity between the fields, the fields whose first similarity is greater than the corresponding threshold are merged into one field; by calculating the data fingerprint of each record, for each record, based on the data fingerprint of each record, the second similarity between the record and other records is calculated, and one record is selected from the records whose second similarity is greater than the corresponding threshold and the other records are deleted to remove redundant data; When the data value corresponding to a field is missing, the data value is automatically inferred or supplemented manually.

3. A method for predicting intraoperative hypertension according to claim 2, characterized in that: The step of calculating the data fingerprint of each record, calculating the second similarity between the record and other records based on the data fingerprint of each record, selecting a record from the records whose second similarity is greater than a corresponding threshold and deleting other records to remove redundant data includes: Step S11, setting the weight of each field; Step S12: For each record, obtain the Hash value of the data value corresponding to each field of the record through the MD5 function, and obtain the data fingerprint of the record based on the Hash value of the data value corresponding to each field and its weight; Step S13: For each record, calculate the Hamming distance between the data fingerprint of the record and the data fingerprints of other records to obtain a second similarity, and select the record with the least number of missing data values ​​corresponding to the field from the records whose second similarity is greater than the corresponding threshold; if there are multiple records with the least number of missing data values ​​corresponding to the fields, select the record with the most recent time and delete the other records.

4. A method for predicting intraoperative hypertension according to claim 3, characterized in that: After encoding the Hash value of the data value corresponding to each field, multiply each bit of the Hash value by its corresponding weight to obtain a weighted Hash value; The weighted Hash value of each field is accumulated according to the corresponding bit to obtain the SimHash value of the record; The SimHash value is converted into binary to obtain the data fingerprint of the record.

5. A method for predicting intraoperative hypertension according to claim 3, characterized in that: The second similarity calculation is performed by the following method: R=1-(H(K1,K2) / W), Among them, R is the similarity, K1 and K2 are two records, H(K1, K2) means calculating the Hamming distance between K1 and K2, and W is the number of bits of the data fingerprint.

6. A method for predicting intraoperative hypertension according to claim 1, characterized in that: The construction rule base includes: For the data value corresponding to each field, a mapping rule is defined based on a regular expression, and the data value is mapped into a standardized expression based on the mapping rule; Building a list based on the standardized expression and corresponding fields; The mapping rules and the list are stored in a database as a rule base.

7. A method for predicting intraoperative hypertension according to claim 1, characterized in that: The selecting of influencing features from the feature variables according to correlation coefficient calculation and expert judgment includes: The characteristic variable is divided into a first characteristic variable and a second characteristic variable, and the correlation coefficient between the data value corresponding to the first characteristic variable and the data value corresponding to the target variable is calculated, and the first characteristic variable corresponding to the correlation coefficient greater than the threshold is selected as the first influencing feature; and the relevant characteristic variable is selected from the second characteristic variable as the second influencing feature according to the expert judgment; The first characteristic variables include age, BMI index, smoking, alcoholism, sodium intake, blood sugar, blood lipids, history of hypertension, history of diabetes, family history of hypertension, and heart rate; the second characteristic variables include use of pressor drugs, use of hormonal drugs, use of non-steroidal anti-inflammatory drugs, use of oral contraceptives, use of antidepressants, and use of anti-tumor drugs.

8. A method for predicting intraoperative hypertension according to claim 7, characterized in that: The correlation coefficient between the data value corresponding to the first feature variable and the data value corresponding to the target variable is calculated by the following formula: Among them, λ i is the correlation coefficient between the data value corresponding to the i-th first feature variable and the data value corresponding to the target variable, F i is the data value corresponding to the first characteristic variable of the i-th order, T is the data value corresponding to the target variable, cov(F i ,T) is F i Covariance with T, σF i F i is the standard deviation of , σT is the standard deviation of T.

9. A method for predicting intraoperative hypertension according to claim 1, characterized in that: The intraoperative hypertension prediction model includes an input layer, a linear combination layer and an output layer; the input layer is used to receive the characteristic values ​​of the samples; the linear combination layer includes a linear combination unit and a mapping function, the linear combination unit is used to perform weighted summation on the input characteristic values, and the mapping function is used to map the output of the linear combination unit to the interval (0,1); the output layer is used to output the predicted probability of the patient developing intraoperative hypertension and the corresponding prediction label.

10. A method for predicting intraoperative hypertension according to claim 9, characterized in that: The predicted probability of intraoperative hypertension in patients was obtained by the following formula: Among them, P i represents the predicted probability of intraoperative hypertension in the ith patient; exp represents the exponential function; W ij is the weight of the jth influencing feature of the i-th patient; X ij is the data value corresponding to the jth influencing feature of the ith patient; n is the number of influencing features, that is, the sum of the first influencing feature, the second influencing feature and the number of target variables; W0 is the bias.

Images