Adverse drug response prediction system using skin electronic medical record

Abstract

The invention relates to the technical field of medical information, in particular to a system for predicting adverse drug response by using electronic skin medical records, which comprises the following steps of: acquiring the electronic skin medical records of a plurality of historical patients, analyzing the electronic skin medical records, and determining the basic correlation degree and effective coefficient of each feature; determining the correlation degree of each feature, and determining the reference weight of a plurality of decision trees obtained by training the electronic medical records of the skin of the historical patient based on the correlation degree; and inputting the skin electronic medical record of the patient to be predicted into a plurality of decision trees, and weighting the output results of the decision trees by using the reference weight so as to obtain the prediction result of the adverse drug reaction of the patient to be predicted. According to the method, the reference weight of the decision tree is adaptively determined, so that the prediction accuracy of adverse drug reaction of the to-be-predicted patient is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and particularly to an adverse drug reaction prediction system using skin electronic medical records. Background Art

[0002] Adverse drug reaction, namely adverse drug reaction (ADR), refers to an adverse reaction that is irrelevant to the treatment purpose and has a causal relationship with the drug application during the process of treating with drugs at the prescribed dose. Common skin adverse drug reactions include drug eruption, photosensitivity reaction, pigmentation, alopecia, etc. During the drug treatment of skin disease patients, due to patient individual factors (such as allergy, heredity, environment), drug interactions (such as when antibiotics and antifungal drugs are used in combination, it may enhance skin allergic reaction or photosensitivity reaction), and the drug itself being irritating, etc., it may cause patients to have adverse drug reactions. When doctors formulate drug treatment plans for skin disease patients, predicting adverse drug reactions in skin disease patients helps to remind doctors of the possible adverse reaction situations in the drugs, so as to assist doctors in adjusting the drug treatment plan or informing patients of the drug adverse reaction situations.

[0003] In the prior art, when predicting adverse drug reactions in skin disease patients, a classification model can be constructed based on machine learning methods such as random forest and support vector machine, and the content in the skin electronic medical records of patients is used to train the classification model, and the trained classification model is used to predict the probability of a patient having an adverse reaction when using a certain drug. Among them, when using a random forest to construct a classification model for adverse drug reaction prediction, the output results of all decision trees in the random forest are averaged to obtain the final prediction result. However, since each decision tree in the random forest uses a part of the features randomly selected from all the features of the patient, and the degree of association between each feature and the adverse reaction is different, the influence degree of the output results of the constructed different decision trees on the prediction result is different. This leads to a large prediction error in directly averaging the output results of all decision trees in the random forest to obtain the final prediction result in the prior art, and ultimately affects the accuracy of predicting adverse drug reactions in skin disease patients. Summary of the Invention

[0004] The purpose of the present invention is to provide an adverse drug reaction prediction system using skin electronic medical records, which is used to solve the problem of poor accuracy in predicting adverse drug reactions in existing skin disease patients.

[0005] To solve the above technical problems, in the first aspect, the present invention provides an adverse drug reaction prediction system using skin electronic medical records, and the system includes: A data acquisition module for acquiring the electronic skin medical records of a number of historical patients. The electronic skin medical records include diagnosis information, and some of the electronic skin medical records also include follow-up visit information. The diagnosis information and the follow-up visit information include the occurrence of various features and the occurrence of adverse drug reactions in patients. The various features include patient symptoms and drug ingredients used in treatment plans; A basic correlation degree acquisition module for statistically analyzing the occurrence of each feature in the diagnosis information of all historical patients and the occurrence of adverse drug reactions in patients, and determining the basic correlation degree of each feature by combining the difference in the occurrence of each feature in the diagnosis information and the follow-up visit information of the same historical patient and the difference in the occurrence of adverse drug reactions in patients; An effective coefficient acquisition module for training a number of decision trees using the electronic skin medical records of the historical patients, and determining the effective coefficient of the basic correlation degree of each feature according to the distribution of historical patients with adverse drug reactions before and after the target item feature divides the parent node in the decision tree and the number of layers of the parent node divided by each feature in the decision tree; An association degree acquisition module for determining the association degree of each feature according to the basic correlation degree of each feature and its effective coefficient; A weight acquisition module for determining the reference weight of the decision tree according to the association degree of all item features included in the decision tree; A prediction module for inputting the electronic skin medical record of the patient to be predicted into the number of decision trees, and weighting the output result of the decision tree using the reference weight to obtain the prediction result of the occurrence of adverse drug reactions in the patient to be predicted.

[0006] Combined with the first aspect above, in some possible implementation manners, the basic correlation degree acquisition module includes: A frequency difference acquisition unit for statistically analyzing the occurrence of the target item feature in the diagnosis information of all historical patients and the occurrence of adverse drug reactions in patients, and determining the frequency difference of the target item feature, where the frequency difference reflects the frequency difference of adverse drug reactions in patients when the target item feature appears or not; A change coefficient acquisition unit for determining the change coefficient of the target item feature according to the total number of historical patients in which there are differences in whether the target item feature appears in the diagnosis information and the follow-up visit information in the electronic skin medical record of the same historical patient and there are also differences in whether adverse drug reactions occur in patients; A basic correlation degree acquisition unit for determining the basic correlation degree of the target item feature according to the difference between the frequency difference of the target item feature and the frequency differences of other item features and the change coefficient of the target item feature.

[0007] Combined with the first aspect above, in some possible implementation manners, the frequency difference obtaining unit includes: A first frequency determining unit, configured to determine a ratio of the number of all historical patients in the diagnostic information who have the target item feature and have adverse drug reactions to the number of all historical patients in the diagnostic information who have the target item feature, so as to obtain a first frequency; A second frequency determining unit, configured to determine a ratio of the number of all historical patients in the diagnostic information who do not have the target item feature and have adverse drug reactions to the number of all historical patients in the diagnostic information who do not have the target item feature, so as to obtain a second frequency; A frequency difference determining unit, configured to determine an absolute value of a difference between the first frequency and the second frequency, so as to obtain a frequency difference of the target item feature.

[0008] Combined with the first aspect above, in some possible implementation manners, the basic correlation degree obtaining unit includes: An average frequency difference determining unit, configured to determine an average frequency difference according to an overall distribution level of frequency differences of all item features; A relative frequency difference value determining unit, configured to determine a ratio of the frequency difference of the target item feature to the average frequency difference, so as to obtain a relative frequency difference value; A relative frequency difference value increment determining unit, configured to determine a product of the relative frequency difference value and a variation coefficient of the target item feature, so as to obtain a relative frequency difference value increment; A basic correlation degree determining unit, configured to perform normalization processing on a sum value of the relative frequency difference value and the relative frequency difference value increment, so as to obtain a basic correlation degree of the target item feature.

[0009] Combined with the first aspect above, in some possible implementation manners, the effective coefficient obtaining module includes: A partition evaluation obtaining unit, configured to determine a partition evaluation of the target item feature for the decision tree according to a difference between a ratio of historical patients with adverse drug reactions in each pair of sibling nodes obtained after the target item feature partitions a parent node in the decision tree and the ratio of historical patients with adverse drug reactions in the parent node, and a difference between ratios of historical patients with adverse drug reactions in each sibling node of each pair of sibling nodes; A partition efficiency obtaining unit, configured to determine a partition efficiency of the target item feature according to the partition evaluation of the target item feature for the decision tree and a layer number of the parent node partitioned by the target item feature in the decision tree; An effective coefficient obtaining unit, configured to determine an effective coefficient of the basic correlation degree of the target item feature according to a difference between the partition efficiency of the target item feature and the partition efficiency of other item features.

[0010] In combination with the above first aspect, in some possible implementation manners, the division evaluation obtaining unit includes: A first ratio difference determining unit, configured to determine a difference between a maximum value of the historical patient ratio of patients with adverse drug reactions in each pair of sibling nodes obtained after dividing the parent node by the target item feature in the decision tree and the historical patient ratio of patients with adverse drug reactions in the parent node, so as to obtain a first ratio difference; A second ratio difference determining unit, configured to determine a difference between the historical patient ratios of patients with adverse drug reactions in each pair of sibling nodes obtained after dividing the parent node by the target item feature in the decision tree, so as to obtain a second ratio difference; A division evaluation determining unit, configured to determine an overall distribution level of the product values of the first ratio difference and the second ratio difference corresponding to each pair of sibling nodes obtained after dividing the parent node by the target item feature in the decision tree, and determine the division evaluation of the target item feature on the decision tree.

[0011] In combination with the above first aspect, in some possible implementation manners, the division efficiency obtaining unit includes: A first division efficiency determining unit, configured to determine a product of the division evaluation of the target item feature on the decision tree and the number of layers of the parent node divided by the target item feature in the decision tree, so as to obtain the division efficiency of the target item feature on a single decision tree; A second division efficiency determining unit, configured to determine an overall distribution level of the division efficiencies of the target item feature on all decision trees, and determine the division efficiency of the target item feature.

[0012] In combination with the above first aspect, in some possible implementation manners, the effective coefficient obtaining unit includes: An average division efficiency determining unit, configured to determine an average division efficiency according to the overall distribution level of the division efficiencies of all item features; A division efficiency difference determining unit, configured to determine a product value of the average division efficiency and a set effective threshold, and further determine a difference between the division efficiency of the target item feature and the product value, so as to obtain a division efficiency difference; An effective coefficient determining unit, configured to input the division efficiency difference into a ReLU function, and output an effective coefficient of the basic correlation degree of the target item feature by the ReLU function.

[0013] In combination with the above first aspect, in some possible implementation manners, the correlation degree obtaining module includes: A correlation degree growth value obtaining unit, configured to determine a product of the basic correlation degree of each item feature and its effective coefficient, so as to obtain a correlation degree growth value; An association degree acquisition unit, configured to perform normalization processing on the sum value of the basic association degree of each feature and the association degree growth value, so as to obtain the association degree of each feature.

[0014] Combined with the first aspect above, in some possible implementation manners, the weight acquisition module includes: An average association degree acquisition unit, configured to determine the average association degree corresponding to the decision tree according to the overall distribution level of the association degrees of all features included in the decision tree; A reference weight acquisition unit, configured to perform normalization processing on the average association degree corresponding to the decision tree to obtain the reference weight of the decision tree, and the cumulative value of the reference weights of all the decision trees is equal to the value 1.

[0015] To solve the above technical problems, in a second aspect, the present invention further provides an adverse drug reaction prediction method using a skin electronic medical record. The method includes: Obtain the skin electronic medical records of several historical patients. The skin electronic medical records include diagnosis information, and some of the skin electronic medical records also include follow-up visit information. The diagnosis information and the follow-up visit information include the occurrence of each feature and the occurrence of adverse drug reactions in the patients. The features include patient symptoms and drug ingredients used in the treatment plan; Statistically analyze the occurrence of each feature in the diagnosis information of all historical patients and the occurrence of adverse drug reactions in the patients, and combine the difference in the occurrence of each feature in the diagnosis information and the follow-up visit information of the same historical patient and the difference in the occurrence of adverse drug reactions in the patients to determine the basic association degree of each feature; Train several decision trees using the skin electronic medical records of the historical patients. According to the distribution of historical patients with adverse drug reactions before and after the target item feature divides the parent node in the decision tree, and the number of layers of the parent node divided by each feature in the decision tree, determine the effective coefficient of the basic association degree of each feature; Determine the association degree of each feature according to the basic association degree of each feature and its effective coefficient; Determine the reference weight of the decision tree according to the association degrees of all features included in the decision tree; Input the skin electronic medical record of the patient to be predicted into the several decision trees, and use the reference weight to weight the output result of the decision tree, so as to obtain the prediction result of the occurrence of adverse drug reactions in the patient to be predicted.

[0016] To solve the above technical problems, in a third aspect, the present invention further provides an adverse drug reaction prediction device using a skin electronic medical record, including a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, so that the device executes the module implementation steps in the above first aspect or any possible implementation manner of the first aspect.

[0017] To solve the above technical problems, in a fourth aspect, the present invention further provides a computer program product, which includes: computer program code, when the computer program code runs on a computer, it causes the computer to execute the module implementation steps in the above first aspect or any possible implementation manner of the first aspect.

[0018] To solve the above technical problems, in a fifth aspect, the present invention further provides a computer-readable storage medium, which stores computer program code, and when the computer program code runs on a computer, it causes the computer to execute the module implementation steps in the above first aspect or any possible implementation manner of the first aspect.

[0019] The present invention has the following beneficial effects: By obtaining the skin electronic medical records of a number of historical patients, the skin electronic medical records include diagnosis information, and some skin electronic medical records also include follow-up visit information. The diagnosis information and follow-up visit information include the occurrence of various features and the occurrence of adverse drug reactions in patients. Each feature includes the patient's symptoms and the drug components used in the treatment plan; By statistically analyzing the occurrence of each feature in the diagnosis information of all historical patients and the occurrence of adverse drug reactions in patients, and combining the difference in the occurrence of each feature in the diagnosis information and the follow-up visit information of the same historical patient and the difference in the occurrence of adverse drug reactions in patients, the basic correlation degree of each feature is determined; At the same time, a number of decision trees are trained using the skin electronic medical records of historical patients. According to the distribution of historical patients with adverse drug reactions before and after the target item feature divides the parent node in the decision tree, and the number of layers of the parent node divided by each feature in the decision tree, the effective coefficient of the basic correlation degree of each feature is determined; Based on the basic correlation degree of each feature and its effective coefficient, the correlation degree of each feature is determined, and based on this correlation degree, the reference weights of a number of decision trees trained using the skin electronic medical records of historical patients are determined; The skin electronic medical record of the patient to be predicted is input into a number of decision trees, and the output results of the decision trees are weighted using the reference weights, so as to obtain the prediction result of the occurrence of adverse drug reactions in the patient to be predicted. The present invention effectively improves the prediction accuracy of the occurrence of adverse drug reactions in the patient to be predicted by adaptively determining the reference weights of the decision trees. Description of the Drawings

[0020] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0021] Figure 1 It is a schematic structural diagram of a system for predicting adverse drug reactions using a skin electronic medical record according to an embodiment of the present invention; Figure 2 It is a flowchart of the steps of a method for predicting adverse drug reactions using a skin electronic medical record according to an embodiment of the present invention; Figure 3 It is a schematic structural diagram of a decision tree according to an embodiment of the present invention; Figure 4 It is a schematic structural diagram of a device for predicting adverse drug reactions using a skin electronic medical record according to an embodiment of the present invention. Detailed implementation manners

[0022] To clearly illustrate the technical features of this solution, the present invention will be described in detail below through specific implementation manners in combination with the drawings.

[0023] The embodiments of the present invention will be described in more detail below with reference to the drawings. Although some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments described herein. On the contrary, these embodiments are provided to more thoroughly and completely understand the present invention. It should be understood that the drawings and embodiments of the present invention are only for exemplary purposes and are not used to limit the protection scope of the present invention.

[0024] It should be understood that the various steps recorded in the method implementation manners of the present invention can be executed in different orders and / or executed in parallel. In addition, the method implementation manners may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this regard.

[0025] The term "including" and its variations used herein are open-ended, that is, "including but not limited to". The term "based on" is "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0026] It should be noted that the concepts such as "first" and "second" mentioned in the present invention are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0027] In the embodiments of the present invention, although operations or steps are described in a specific order in the drawings, it should not be understood that these operations or steps are required to be performed in the specific order shown or in a serial order, or that all the shown operations or steps are required to be performed to obtain the desired result. In the embodiments of the present invention, these operations or steps can be performed serially; they can also be performed in parallel; or a part of these operations or steps can be performed.

[0028] Meanwhile, it can be understood that the data involved in the technical solution of the present invention (including but not limited to the data itself, the acquisition or use of the data) should comply with the requirements of the corresponding laws, regulations and related provisions. Unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs, and all parameters or indicators in the formulas involved in the present invention are numerical values after normalization that eliminate the influence of dimensions.

[0029] To solve the problem of poor accuracy in predicting adverse drug reactions for existing skin disease patients, an adverse drug reaction prediction system using skin electronic medical records is provided in the embodiments of the present invention. This system is essentially a software system, which is composed of modules that implement corresponding functions, and its corresponding structural schematic diagram is as Figure 1 shown. The core of this system is to implement an adverse drug reaction prediction method using skin electronic medical records. Each module in this system corresponds to each step in the method, and the flowchart corresponding to this method is as Figure 2 shown. The following will introduce each module of this system in detail in combination with the specific steps in this method.

[0030] A data acquisition module, which is used to acquire the skin electronic medical records of a number of historical patients. The skin electronic medical records include diagnosis information, and some of the skin electronic medical records also include follow-up visit information. The diagnosis information and the follow-up visit information include the occurrence of various features and the occurrence of adverse drug reactions in the patients. The various features include patient symptoms and the drug ingredients used in the treatment plan.

[0031] Specifically, in order to facilitate the subsequent construction of an adverse drug reaction prediction model, it is first necessary to obtain a number of historical skin electronic medical record samples. In this embodiment, the skin electronic medical records of a number of historical patients are obtained, and the diagnosis information of the historical patients is included in the skin electronic medical records. The number of historical patients can be reasonably set as needed. For example, it is set to 1000 historical patients. At the same time, if there are follow-up visit records of the historical patients, the follow-up visit information of the historical patients is also included in the obtained skin electronic medical records of these historical patients.

[0032] The occurrence of each feature and the occurrence of adverse drug reactions in the diagnosis information and follow-up visit information included in the skin electronic medical records of a number of historical patients. Among them, each feature includes each feature in the patient's symptoms and drug treatment plan. The features of the patient's symptoms include whether the patient has a certain skin manifestation or is allergic to a certain drug. The skin manifestations include itching, rash, desquamation, exudation, etc.; the features of the drug treatment plan refer to whether a certain drug ingredient is used in the treatment plan. The drug ingredients include chloramphenicol, erythromycin, mupirocin, loratadine, etc. By inputting the features in the diagnosis information and follow-up visit information included in the skin electronic medical records as binary variables, taking whether there is a certain skin manifestation as an example, when there is a certain skin manifestation, the quantification of this feature is "1", and when there is no certain skin manifestation, the quantification of this feature is "0". The same quantification process is performed for whether the patient is allergic to a certain drug and whether a certain drug ingredient is used. If the patient is allergic to a certain drug or a certain drug ingredient is used, the corresponding feature is quantified as "1", otherwise the corresponding feature is quantified as "0". For example, if the patient's medication includes loratadine and does not include mupirocin, the corresponding variable for taking loratadine is quantified as 1, and the corresponding variable for taking mupirocin is quantified as 0.

[0033] In this way, by quantifying the features in the diagnosis information and follow-up visit information included in the skin electronic medical records, the quantification values of the features in the diagnosis information and follow-up visit information included in the skin electronic medical records can be obtained. When the quantification value is 1, it means that the corresponding feature appears, and when the quantification value is 0, it means that the corresponding feature does not appear. It should be understood that the skin electronic medical records of the above-mentioned number of historical patients are all medical record data obtained with the permission of the patients.

[0034] The basic correlation degree acquisition module is used to count the occurrence of each feature in the diagnosis information of all historical patients and the occurrence of adverse drug reactions in the patients, and determine the basic correlation degree of each feature by combining the difference in the occurrence of each feature in the diagnosis information and the follow-up visit information of the same historical patient and the difference in the occurrence of adverse drug reactions in the patients.

[0035] Specifically, the skin electronic medical record data contains a large amount of patient characteristic information, and this information is usually high-dimensional, diverse, and has complex non-linear relationships. By integrating multiple decision trees, random forest can automatically process various features, adapt to the diversity of data, and extract effective feature information for prediction. Therefore, in this embodiment, a prediction model for adverse drug reactions is established based on random forest. In the process of establishing a prediction model for adverse drug reactions based on random forest, several sub-training sets are extracted from the training set, and each sub-training set is used to train a decision tree. After several decision trees are trained, the current quantity to be predicted is input into each decision tree to obtain several output results, and a combiner is constructed to combine the output results of each decision tree as the output of the final random forest.

[0036] Since each decision tree in the random forest uses a part of the features randomly selected from all the features of the skin disease patients, the random selection method of the random forest will be affected by the corresponding association relationship formed by different features in the skin electronic medical records of the patients and the adverse drug reactions, resulting in the weakening of the feature association relationship in the random division process. By analyzing the association degree between each feature and the occurrence of adverse drug reactions based on the skin electronic medical records of all historical patients, the reference weight of each decision tree is determined according to the features included in each decision tree, and the output results of each decision tree are weighted using the reference weight, which can effectively improve the prediction accuracy of the random forest.

[0037] Considering that the skin electronic medical records of skin disease patients contain multiple features, and the manifestations of multiple features may all lead to adverse drug reactions in patients. The manifestations of adverse drug reactions are related to features such as the thickness of the patient's skin surface layer and ultraviolet sensitivity. Different features, such as different skin property features, have different degrees of association with the adverse drug reactions manifested. When a patient has symptoms of skin desquamation or exfoliation, the skin surface layer of the patient becomes thinner and the skin is more sensitive. When the skin is stimulated by the outside world, it is easy for the patient to have adverse drug reactions such as rashes. The difference in the statistical results in the skin electronic medical records of all historical patients can reflect the association degree between different features and adverse drug reactions.

[0038] Further, the above basic association degree acquisition module includes: A frequency difference acquisition unit, which is used to count the occurrence of the target item feature in the diagnosis information of all historical patients and the occurrence of adverse drug reactions in the patients, and determine the frequency difference of the target item feature. The frequency difference reflects the frequency difference of the occurrence of adverse drug reactions in patients when the target item feature appears or not; A variation coefficient acquisition unit is configured to determine the variation coefficient of the target item feature according to the total number of historical patients in the skin electronic medical record of the same historical patient, where there are differences in whether the target item feature appears in the diagnosis information and the follow-up visit information, and there are also differences in whether the patient has an adverse drug reaction. A basic correlation degree acquisition unit is configured to determine the basic correlation degree of the target item feature according to the difference between the frequency difference of the target item feature and the frequency difference of other item features, and the variation coefficient of the target item feature.

[0039] Specifically, since the different features in the skin electronic medical record of patients with skin diseases have different degrees of association with the occurrence of adverse drug reactions, some features will trigger adverse drug reactions only under certain specific circumstances. For example, some drugs that cause photosensitivity usually trigger adverse reactions under high-intensity ultraviolet radiation, while drugs such as tetracycline can activate the body's immune system, causing the human immune system to have an allergic reaction to the drug and easily triggering adverse reactions. By statistically calculating the probability of the presence or absence of adverse drug reactions in each feature of all historical patients, the basic correlation degree of each feature can be obtained, which is used to reflect the occurrence frequency of adverse drug reactions under different manifestation conditions of each feature.

[0040] When the difference in the occurrence frequency of adverse drug reactions is large when a certain feature is present or absent, the basic correlation degree of this feature is greater, indicating that this feature is more likely to be associated with the manifestation of adverse drug reactions. If the presence or absence of a certain feature has little impact on the occurrence frequency of adverse drug reactions, the basic correlation degree of this feature is smaller.

[0041] Furthermore, the above frequency difference acquisition unit includes: a first frequency determination unit configured to determine the ratio of the number of all historical patients in the diagnosis information who have the target item feature and the patient has an adverse drug reaction to the number of all historical patients in the diagnosis information who have the target item feature, to obtain a first frequency; a second frequency determination unit configured to determine the ratio of the number of all historical patients in the diagnosis information who do not have the target item feature and the patient has an adverse drug reaction to the number of all historical patients in the diagnosis information who do not have the target item feature, to obtain a second frequency; a frequency difference determination unit configured to determine the absolute value of the difference between the first frequency and the second frequency, so as to obtain the frequency difference of the target item feature.

[0042] In this embodiment, taking the th feature as the target item feature as an example, the frequency difference of the th feature ; where: represents the number of historical patients in the diagnosis information who have the The number of all historical patients with the item feature and adverse drug reactions occurred in the patients, that is, the number of all historical patients with adverse drug reactions when the quantification value of the item feature takes 1; represents the number of all historical patients with the item feature in the diagnostic information, that is, the number of all historical patients with the quantification value of the item feature taking 1; represents the number of all historical patients with the item feature not in the diagnostic information and adverse drug reactions occurred in the patients, that is, the number of all historical patients with adverse drug reactions when the quantification value of the item feature takes 0;

[0043] In the above formula, when the difference between the first frequency and the second frequency is large, it indicates that the frequency difference of adverse drug reactions in patients when the item feature appears or not is large, then the item feature is more likely to be associated with the manifestation of adverse drug reactions.

[0044] Since the skin electronic medical records of historical patients will show a phased increase in content with the progress of follow-up visits, including diagnostic information, medication information, etc., and its general situation is to add or subtract the categories of drug types or change the dosage according to the manifestation of skin drug reactions, then the diagnostic features of the patients also change accordingly. At this time, if the situation of adverse drug reactions also changes after several features of the same patient change, such as from none to some or from some to none, it means that these features are more relevant to adverse drug reactions.

[0045] In this embodiment, taking the item feature as the target item feature as an example, the change coefficient of the item feature is calculated as follows: ; where: represents the total number of historical patients with follow-up visit information in the skin electronic medical records; represents the The differential marker value of whether a patient has an adverse drug reaction in the diagnostic information and follow-up visit information in the skin electronic medical record of historical patients. When the patient has an adverse drug reaction in both the diagnostic information and the follow-up visit information, or does not have an adverse drug reaction in both, the differential marker value ; represents the differential marker value of whether the th feature in the diagnostic information and follow-up visit information in the skin electronic medical record of the th historical patient including the follow-up visit information in the skin electronic medical record appears. When the th feature appears in both the diagnostic information and the follow-up visit information, or does not appear in both, the differential marker value ; otherwise, the differential marker value ; represents the linear normalization function.

[0046] In the above formula, when more historical patients show that after the th feature in the diagnostic information and the follow-up visit information changes, the corresponding situation of having an adverse drug reaction also changes, it indicates that the th feature is more likely to be associated with the manifestation of adverse drug reactions.

[0047] The frequency difference reflects the correlation between the presence or absence of each feature in the diagnostic information of all skin electronic medical records and the presence or absence of adverse drug reactions. If the situation of patients having adverse drug reactions in the diagnosis and follow-up visits changes, then the features with different manifestations in the diagnosis and follow-up visits are very likely to have a high degree of association with adverse drug reactions. Therefore, by combining the frequency difference and the follow-up visit change situation, the basic association degree of each feature with adverse drug reactions can be obtained.

[0048] Furthermore, the basic association degree acquisition unit includes: an average frequency difference determination unit for determining the average frequency difference according to the overall distribution level of the frequency differences of all features; a relative frequency difference determination unit for determining the ratio of the frequency difference of the target feature to the average frequency difference to obtain the relative frequency difference value; a relative frequency difference increment determination unit for determining the product of the relative frequency difference value and the change coefficient of the target feature to obtain the relative frequency difference increment; a basic association degree determination unit for normalizing the sum of the relative frequency difference value and the relative frequency difference increment to obtain the basic association degree of the target feature.

[0049] In this embodiment, taking the th feature as an example of the target feature, the formula for calculating the basic association degree of the th feature is: ; Wherein: represents the coefficient of variation of the th feature; represents the frequency difference of the th feature; represents the average frequency difference, that is, the average value of the frequency differences of all item features; represents the linear normalization function.

[0050] In the above formula, by calculating the ratio of the frequency difference of the th feature to the average frequency difference, the relative value of the frequency difference is obtained, and the product of the relative value of the frequency difference and the coefficient of variation of the th feature is determined to obtain the relative value increment of the frequency difference . Furthermore, by using the function to normalize the sum value of the relative value of the frequency difference and the relative value increment of the frequency difference , the basic correlation degree is finally obtained. When the relative value of the frequency difference of the th feature is larger, and the coefficient of variation of the th feature is larger, it indicates that the th feature is more likely to be associated with the manifestation of adverse drug reactions, and the corresponding basic correlation degree has a larger value.

[0051] In the above manner, the basic correlation degrees of all item features in the skin electronic medical records of all historical patients can be determined.

[0052] The effective coefficient acquisition module is used to train a number of decision trees by using the skin electronic medical records of the historical patients, and determine the effective coefficient of the basic correlation degree of each feature according to the distribution of historical patients with adverse drug reactions before and after the target item feature divides the parent node in the decision tree, and the number of layers of the parent node divided by each feature in the decision tree.

[0053] Specifically, according to the acquisition process of the basic correlation degrees of all the above item features, the basic correlation degree of each feature is obtained based on the statistical results of the appearance frequencies of each feature and adverse drug reactions among all historical patients. The basic correlation degree can reflect the correlation between the feature and adverse drug reactions to a certain extent. However, since each historical patient has multiple features at the same time, some features may not affect the occurrence of adverse drug reactions, but if their appearance frequencies are high and they appear together with features with high correlation in most medical records, it will also lead to a high basic correlation degree. Therefore, the basic correlation degree obtained only based on the overall statistical results cannot be directly used as the degree of correlation between each feature and adverse drug reactions, and further analysis is required to accurately determine the degree of correlation of each feature.

[0054] Considering that when training several decision trees using the skin electronic medical records of historical patients, the data subsets used are different, that is, each decision tree uses different samples and features, which can effectively reduce the dependence of the prediction model on a single data sample. Therefore, analyzing within each decision tree can discuss the patient classification situation when some features are the same and some features are consistent at a smaller scale.

[0055] Furthermore, the above-mentioned effective coefficient acquisition module includes: A division evaluation acquisition unit, which is used to determine the division evaluation of the target item feature for the decision tree according to the difference between the proportion of historical patients with adverse drug reactions in each pair of sibling nodes obtained by dividing the parent node according to the target item feature in the decision tree and the proportion of historical patients with adverse drug reactions in the parent node, and the difference between the proportion of historical patients with adverse drug reactions in each sibling node of each pair of sibling nodes; A division efficiency acquisition unit, which is used to determine the division efficiency of the target item feature according to the division evaluation of the target item feature for the decision tree and the layer number of the parent node divided by the target item feature in the decision tree; An effective coefficient acquisition unit, which is used to determine the effective coefficient of the basic correlation degree of the target item feature according to the difference between the division efficiency of the target item feature and the division efficiency of other item features.

[0056] Specifically, use the information in the skin electronic medical records of historical patients to train several decision trees. Denote 80% of the total number of skin electronic medical records of all historical patients as the training set size N, and the remaining 20% of the skin electronic medical records except the training set as the test set, and denote the number of categories of all item features as M.

[0057] Randomly draw a sub-training set of size n with replacement in the training set N, and randomly draw m item features without replacement from all features M. Use the obtained sub-training set of size n and m item features to train a decision tree. The output result of each decision tree is whether there is an adverse drug reaction. If there is an adverse drug reaction, the decision tree outputs "1", otherwise it outputs "0". Preset the number of decision trees R, and repeat the above process of training decision trees R times to obtain a total of R decision trees. Since the process of training decision trees belongs to the prior art, it will not be elaborated here.

[0058] Since each layer of the decision tree represents dividing a set using an item feature, such as Figure 3As shown, Y represents a historical patient who has had an adverse drug reaction, and X represents a historical patient who has not had an adverse drug reaction. As the number of layers of the decision tree increases, the features between sibling nodes become more consistent, that is, they have shown consistency in multiple features. For example, among the historical patients within the same node in the 4th layer, it means that the manifestations of the previously divided 3 features are all consistent. For multiple historical patients that are more consistent, if the frequency of adverse drug reactions varies greatly after being divided by a certain feature, it indicates that the correlation between this feature and adverse drug reactions is greater.

[0059] Furthermore, the above-mentioned division evaluation acquisition unit includes: a first ratio difference determination unit, which is used to determine the difference between the maximum value of the proportion of historical patients with adverse drug reactions in each pair of sibling nodes obtained after dividing the parent node by the target item feature in the decision tree and the proportion of historical patients with adverse drug reactions in the parent node, to obtain the first ratio difference; a second ratio difference determination unit, which is used to determine the difference between the proportions of historical patients with adverse drug reactions in each pair of sibling nodes obtained after dividing the parent node by the target item feature in the decision tree, to obtain the second ratio difference; a division evaluation determination unit, which is used to determine the overall distribution level of the product values of the first ratio difference and the second ratio difference corresponding to each pair of sibling nodes obtained after dividing the parent node by the target item feature in the decision tree, and determine the division evaluation of the target item feature for the decision tree.

[0060] In this embodiment, taking the th feature as the target item feature as an example, for the th decision tree, if the th feature appears in the th decision tree, the division evaluation of the th feature for the th decision tree ; Where: represents the number of pairs of sibling nodes obtained after dividing the parent node by the th feature in the th decision tree. As shown in Figure 3 , the th feature corresponds to 2 pairs of sibling nodes. Each pair of sibling nodes refers to 2 nodes with the same parent node. The first pair of sibling nodes consists of nodes YX and YY, and the second pair of sibling nodes consists of nodes Y and XX; represents the maximum value of the proportion of historical patients with adverse drug reactions in the th pair of sibling nodes obtained after dividing the parent node by the th feature in the th decision tree; Indicates the th decision tree, the th feature, after partitioning the parent node, the proportion of historical patients with adverse drug reactions in the parent node of the pair of sibling nodes obtained; Indicates the th decision tree, the th feature, after partitioning the parent node, the difference between the proportions of historical patients with adverse drug reactions in the pair of sibling nodes obtained, where the difference refers to the absolute value of the difference between the proportions of historical patients corresponding to the two nodes in the pair of sibling nodes.

[0061] In the above formula, for each decision tree with the th feature, when the maximum value of the proportion of historical patients with adverse drug reactions in each pair of sibling nodes obtained after partitioning the parent node by the th feature in each decision tree, and the proportion of historical patients with adverse drug reactions in the parent node has a greater difference, and the difference between the proportions of historical patients with adverse drug reactions in each pair of sibling nodes is greater, it indicates that the th feature has a greater degree of association with adverse drug reactions.

[0062] Combining the partitioning evaluation of each feature for each decision tree and the corresponding layer number of each feature in each decision tree, the partitioning efficiency of each feature is determined. When the corresponding layer number of each feature in each decision tree is larger, it means that there are more layers of features showing consistency before partitioning by this feature in each decision tree. Then, the frequency difference calculated from the partitioning result of this feature excludes the influence of these consistent features.

[0063] Furthermore, the above partitioning efficiency obtaining unit includes: a first partitioning efficiency determining unit, which is used to determine the product of the partitioning evaluation of the target item feature for the decision tree and the layer number of the parent node partitioned by the target item feature in the decision tree, to obtain the partitioning efficiency of the target item feature for a single decision tree; a second partitioning efficiency determining unit, which is used to determine the overall distribution level of the partitioning efficiency of the target item feature for all decision trees, and determine the partitioning efficiency of the target item feature.

[0064] In this embodiment, taking the th feature as the target item feature as an example, the formula for the partitioning efficiency of the th feature is: ; Where: represents the number of all decision trees containing the th feature; Indicates the layer number of the parent node divided by the th feature in the th decision tree; Indicates the th feature's partitioning evaluation of the th decision tree; Indicates the linear normalization function.

[0065] When the partitioning efficiency of a certain feature is low while the basic correlation degree is relatively high, it means that the high basic correlation degree of this feature is caused by other features that appear simultaneously, and this feature itself is not directly highly correlated with adverse drug reactions. For example, skin itching is a common symptom that often appears in various skin diseases. Skin itching does not directly cause adverse reactions, but may be the adverse reactions caused by other symptoms or treatment drugs of a certain disease with the feature of skin itching. When analyzing the correlation degree between this feature and adverse reactions only through the basic correlation degree, the result obtained may be high. Therefore, by combining the differences in the partitioning efficiencies of each feature, the effective coefficient of the basic correlation degree of each feature is determined. Furthermore, by combining the basic correlation degree of each feature and its effective coefficient, the correlation degree of each feature can be accurately obtained.

[0066] Furthermore, the above-mentioned effective coefficient acquisition unit includes: an average partitioning efficiency determination unit for determining the average partitioning efficiency according to the overall distribution level of the partitioning efficiencies of all features; a partitioning efficiency difference determination unit for determining the product of the average partitioning efficiency and a set effective threshold, and then determining the difference between the partitioning efficiency of the target feature and the product value to obtain the partitioning efficiency difference; an effective coefficient determination unit for inputting the partitioning efficiency difference into the ReLU function, and outputting the effective coefficient of the basic correlation degree of the target feature by the ReLU function.

[0067] In this embodiment, the preset effective threshold = 0.5 to limit the lower limit of the partitioning efficiency to avoid the situation where the value of the basic correlation degree is too high when the partitioning efficiency of a certain feature is too low. Taking the th feature as an example of the target feature, the calculation formula for the effective coefficient of the basic correlation degree of the th feature is: ; In the formula: Indicates the partitioning efficiency of the th feature; Indicates the preset effective threshold; Indicates the average value of the partitioning efficiencies of all features, that is, the average partitioning efficiency; Denote the ReLU function. When the input quantity is less than 0, the output quantity of the ReLU function is 0. When the input quantity is greater than or equal to 0, the output quantity of the ReLU function is the input quantity.

[0068] In the above manner, the effective coefficient of the basic correlation degree of all item features in the skin electronic medical records of all historical patients can be determined.

[0069] The correlation degree acquisition module is used to determine the correlation degree of each item feature according to the basic correlation degree of each item feature and its effective coefficient.

[0070] Specifically, by comprehensively considering the basic correlation degree of each item feature and its effective coefficient, the correlation degree of each item feature can be determined. The higher the basic correlation degree and its effective coefficient, the higher the correlation degree between the corresponding feature and the drug adverse reaction.

[0071] Furthermore, the correlation degree acquisition module includes: a correlation degree growth value acquisition unit, which is used to determine the product of the basic correlation degree of each item feature and its effective coefficient to obtain a correlation degree growth value; a correlation degree acquisition unit, which is used to perform normalization processing on the sum value of the basic correlation degree of each item feature and the correlation degree growth value to obtain the correlation degree of each item feature.

[0072] In this embodiment, taking the th item feature as the target item feature as an example, the correlation degree of the th item feature is calculated as follows: where: represents the effective coefficient of the basic correlation degree of the th item feature; represents the basic correlation degree of the th item feature; represents the linear normalization function.

[0073] In the above formula, by calculating the product of the basic correlation degree of the th item feature and its effective coefficient, the correlation degree growth value is obtained. Then, the sum value of the basic correlation degree of the th item feature and the correlation degree growth value is normalized, so as to obtain the correlation degree of the th item feature.

[0074] In the above manner, the degree of association of all item features in the skin electronic medical records of all historical patients can be determined. The basic degree of association of all item features is obtained through the statistical results. By determining the partitioning efficiency of all item features, the effective coefficient of the basic degree of association of all item features is determined, and the basic degree of association is corrected using this effective coefficient, so as to finally obtain the degree of association of all item features. Compared with only obtaining the basic degree of association of all item features through the statistical results as the degree of association, calculating the partitioning efficiency of all item features excludes the influence of some other features, making the credibility of obtaining the correlation between the performance changes of each feature and adverse drug reactions higher.

[0075] A weight acquisition module, configured to determine the reference weight of the decision tree according to the degree of association of all item features included in the decision tree.

[0076] Specifically, the above analysis of the degree of association between each feature and adverse drug reactions based on the changes in each feature and the proportion of adverse drug reactions. Since the degrees of association between different features and adverse drug reactions are different, and the features in each decision tree are randomly selected, the reference weights of different decision trees for the output results of the random forest are different.

[0077] Furthermore, the above weight acquisition module includes: An average degree of association acquisition unit, configured to determine the average degree of association corresponding to the decision tree according to the overall distribution level of the degrees of association of all item features included in the decision tree; A reference weight acquisition unit, configured to perform normalization processing on the average degree of association corresponding to the decision tree to obtain the reference weight of the decision tree, and the cumulative value of the reference weights of all the decision trees is equal to the value 1.

[0078] In this embodiment, the average value of the degrees of association of all item features included in each decision tree is determined, so as to obtain the average degree of association corresponding to each decision tree. Calculate the cumulative value of the average degrees of association corresponding to all decision trees, and calculate the ratio of the average degree of association corresponding to each decision tree to the cumulative value of this average degree of association to perform sum normalization on the average degree of association corresponding to each decision tree, and use this ratio as the reference weight of each decision tree. At this time, the cumulative value of the reference weights of all decision trees is equal to the value 1.

[0079] In the above manner, the reference weights of all decision trees can be determined.

[0080] A prediction module, configured to input the skin electronic medical record of the patient to be predicted into the several decision trees, and use the reference weight to weight the output result of the decision tree, so as to obtain the prediction result of the occurrence of adverse drug reactions in the patient to be predicted.

[0081] Specifically, input the skin electronic medical record of the current patient to be predicted into each decision tree, and obtain the output result of each decision tree. The output result takes a value of 0 or 1. A value of 0 indicates that an adverse drug reaction does not occur, and a value of 1 indicates that an adverse drug reaction does not occur. Multiply the output results of all decision trees by the corresponding reference weights to obtain the weighted output results of all decision trees. Based on the weighted output results of all decision trees, combine the weighted output results of each decision tree through a combiner to determine the prediction result of whether the patient to be predicted has an adverse drug reaction. In this embodiment, determine the average value of the weighted output results of all decision trees, and use this average value as the probability that the patient to be predicted has an adverse drug reaction. If this probability is greater than or equal to 0.5, the prediction result of whether the patient to be predicted has an adverse drug reaction is that the patient to be predicted will have an adverse drug reaction; otherwise, the prediction result of whether the patient to be predicted has an adverse drug reaction is that the patient to be predicted will not have an adverse drug reaction. At the same time, mark the features involved in several decision trees with the largest reference weights whose output results are 1 as key features to show the doctor the key features that may cause adverse drug reactions in the current patient, facilitating the doctor to evaluate the possible situation of adverse drug reactions in the patient to be predicted and adjust the treatment plan according to the key features.

[0082] Based on the same inventive concept, an embodiment of the present invention further provides a method for predicting adverse drug reactions using skin electronic medical records, as Figure 2 shown. The method includes: Obtain the skin electronic medical records of several historical patients. The skin electronic medical records include diagnostic information, and some of the skin electronic medical records also include follow-up visit information. The diagnostic information and the follow-up visit information include the occurrence of each feature and the situation of the patient having an adverse drug reaction. Each feature includes the patient's symptoms and the drug ingredients used in the treatment plan; Statistically analyze the occurrence of each feature in the diagnostic information of all historical patients and the situation of the patient having an adverse drug reaction, and combine the difference in the occurrence of each feature in the diagnostic information and the follow-up visit information of the same historical patient and the difference in the situation of the patient having an adverse drug reaction to determine the basic correlation degree of each feature; Train several decision trees using the skin electronic medical records of the historical patients. According to the distribution of historical patients having an adverse drug reaction before and after the target item feature divides the parent node in the decision tree, and the number of layers of the parent node divided by each feature in the decision tree, determine the effective coefficient of the basic correlation degree of each feature; Determine the correlation degree of each feature according to the basic correlation degree of each feature and its effective coefficient; Determine the reference weight of the decision tree according to the correlation degree of all item features included in the decision tree; Input the skin electronic medical record of the patient to be predicted into the several decision trees, and use the reference weights to weight the output results of the decision trees, so as to obtain the prediction result of the adverse drug reaction of the patient to be predicted.

[0083] Based on the same inventive concept, an embodiment of the present invention further provides an adverse drug reaction prediction device using a skin electronic medical record, as Figure 4 shown. The device includes: a memory 401, a processor 402, and a computer program 403 stored in the memory 401 and running on the processor 402. When the processor 402 executes the computer program 403, the system can execute the module implementation steps in any of the foregoing adverse drug reaction prediction systems using a skin electronic medical record.

[0084] The embodiment of the present invention can divide the functions of the device according to the module implementation step examples in the above system. For example, it can correspond to each function module, or integrate two or more functions into one processing module. The above integrated module can be implemented in the form of hardware. It should be noted that the division of modules in this embodiment is illustrative, only a logical function division, and there may be other division methods in actual implementation.

[0085] Based on the same inventive concept, an embodiment of the present invention further provides a computer program product, which includes: computer program code. When the computer program code runs on a computer, the computer is enabled to execute the module implementation steps in any of the foregoing adverse drug reaction prediction systems using a skin electronic medical record.

[0086] Based on the same inventive concept, an embodiment of the present invention further provides a computer-readable storage medium, which stores computer program code. When the computer program code runs on a computer, the computer is enabled to execute the module implementation steps in any of the foregoing adverse drug reaction prediction systems using a skin electronic medical record.

[0087] It should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A system for predicting adverse drug reactions using skin electronic medical records, characterized in that: The system comprises: A data acquisition module is used to acquire skin electronic medical records of several historical patients, wherein the skin electronic medical records include diagnosis information, and some of the skin electronic medical records also include follow-up visit information, wherein the diagnosis information and the follow-up visit information include the occurrence of various characteristics and the occurrence of adverse drug reactions in patients, wherein the various characteristics include patient symptoms and drug ingredients used in treatment plans; A basic correlation degree acquisition module is used to collect statistics on the occurrence of each feature in the diagnosis information of all historical patients and the occurrence of adverse drug reactions in patients, and determine the basic correlation degree of each feature by combining the differences in the occurrence of each feature in the diagnosis information and the follow-up information of the same historical patient and the differences in the occurrence of adverse drug reactions in patients; An effective coefficient acquisition module is used to train a plurality of decision trees using the skin electronic medical records of the historical patients, and determine the effective coefficient of the basic correlation of each feature according to the distribution of historical patients with adverse drug reactions before and after the parent node is divided by the target item feature in the decision tree, and the number of layers of the parent node divided by each feature in the decision tree; A correlation degree acquisition module is used to determine the correlation degree of each feature according to the basic correlation degree and its effective coefficient of each feature; A weight acquisition module, used to determine the reference weight of the decision tree according to the association degree of all item features included in the decision tree; The prediction module is used to input the skin electronic medical records of the patient to be predicted into the several decision trees, and use the reference weights to weight the output results of the decision trees to obtain the prediction results of adverse drug reactions of the patient to be predicted.

2. The system for predicting adverse drug reactions using skin electronic medical records according to claim 1, characterized in that: The basic relevance acquisition module comprises: A frequency difference acquisition unit is used to collect statistics on the occurrence of target item features in the diagnostic information of all historical patients and the occurrence of adverse drug reactions in patients, and determine the frequency difference of the target item features, which reflects the frequency difference of adverse drug reactions in patients when the target item features appear or not; A variation coefficient acquisition unit is used to determine the variation coefficient of the target item feature according to the total number of historical patients in which the diagnosis information in the skin electronic medical record and the follow-up information of the same historical patient have differences in the target item feature, and whether the patient's adverse drug reactions also have differences; The basic relevance acquisition unit is used to determine the basic relevance of the target item feature according to the difference between the frequency difference of the target item feature and the frequency difference of other item features, and the variation coefficient of the target item feature.

3. The system for predicting adverse drug reactions using skin electronic medical records according to claim 2, characterized in that: The frequency difference acquisition unit comprises: A first frequency determination unit is used to determine the ratio of the number of all historical patients in which the target item feature appears in the diagnostic information and the patients have adverse drug reactions to the number of all historical patients in which the target item feature appears in the diagnostic information, to obtain a first frequency; A second frequency determination unit is used to determine the ratio of the number of all historical patients who do not have the target item feature in the diagnostic information and who have adverse drug reactions to the number of all historical patients who do not have the target item feature in the diagnostic information, to obtain a second frequency; The frequency difference determination unit is used to determine the absolute value of the difference between the first frequency and the second frequency, so as to obtain the frequency difference of the target item feature.

4. The system for predicting adverse drug reactions using skin electronic medical records according to claim 2, characterized in that: The basic relevance acquisition unit comprises: An average frequency difference determination unit, used for determining an average frequency difference according to the overall distribution level of the frequency differences of all item features; A frequency difference relative value determination unit, used to determine the ratio of the frequency difference of the target item feature to the average frequency difference, and obtain a frequency difference relative value; A frequency difference relative value increment determination unit, used to determine the product of the frequency difference relative value and the change coefficient of the target item feature to obtain the frequency difference relative value increment; The basic relevance determination unit is used to normalize the sum of the relative frequency difference value and the relative frequency difference value increment, so as to obtain the basic relevance of the target item feature.

5. The system for predicting adverse drug reactions using skin electronic medical records according to claim 1, characterized in that: The effective coefficient acquisition module comprises: A division evaluation acquisition unit, used to determine the division evaluation of the decision tree by the target item feature according to the difference between each pair of sibling nodes obtained after the parent node is divided by the target item feature in the decision tree and the historical proportion of patients with adverse drug reactions in the parent node, and the difference between the historical proportion of patients with adverse drug reactions in each sibling node in each pair of sibling nodes; A division efficiency acquisition unit, used to determine the division efficiency of the target item feature according to the evaluation of the division of the decision tree by the target item feature and the number of layers of the parent node divided by the target item feature in the decision tree; The effective coefficient acquisition unit is used to determine the effective coefficient of the basic correlation degree of the target item feature according to the difference between the division efficiency of the target item feature and the division efficiency of other item features.

6. The system for predicting adverse drug reactions using skin electronic medical records according to claim 5, characterized in that: The division evaluation acquisition unit comprises: A first proportion difference determination unit is used to determine the maximum value of the historical proportion of patients with adverse drug reactions in each pair of brother nodes obtained after the target item feature divides the parent node in the decision tree, and the difference between the historical proportion of patients with adverse drug reactions in the parent node, to obtain a first proportion difference; A second proportion difference determination unit is used to determine the difference between the historical proportions of patients with adverse drug reactions in each pair of brother nodes obtained after the target item feature divides the parent node in the decision tree, to obtain a second proportion difference; A partitioning evaluation determination unit is used to determine the overall distribution level of the product values ​​of the first proportion difference and the second proportion difference corresponding to each pair of brother nodes obtained after the target item feature divides the parent node in the decision tree, and determine the partitioning evaluation of the decision tree by the target item feature.

7. The system for predicting adverse drug reactions using skin electronic medical records according to claim 5, characterized in that: The partitioning efficiency acquisition unit includes: A first division efficiency determination unit is used to determine the product of the division evaluation of the target item feature on the decision tree and the number of layers of the parent node divided by the target item feature in the decision tree, so as to obtain the division efficiency of the target item feature on a single decision tree; The second division efficiency determination unit is used to determine the overall distribution level of the division efficiency of the target item feature for all decision trees, and determine the division efficiency of the target item feature.

8. The system for predicting adverse drug reactions using skin electronic medical records according to claim 5, characterized in that: The effective coefficient acquisition unit comprises: An average partitioning efficiency determination unit, used to determine the average partitioning efficiency according to the overall distribution level of the partitioning efficiency of all item features; A division efficiency difference determination unit, used to determine the multiplication value of the average division efficiency and a set effective threshold, and then determine the difference between the division efficiency of the target item feature and the multiplication value to obtain the division efficiency difference; The effective coefficient determination unit is used to input the division efficiency difference into the ReLU function, and the ReLU function outputs the effective coefficient of the basic correlation degree of the target item feature.

9. The system for predicting adverse drug reactions using skin electronic medical records according to claim 1, characterized in that: The association degree acquisition module includes: A correlation growth value acquisition unit is used to determine the product of the basic correlation of each feature and its effective coefficient to obtain a correlation growth value; The association degree acquisition unit is used to normalize the sum of the basic association degree of each feature and the association degree growth value to obtain the association degree of each feature.

10. The system for predicting adverse drug reactions using skin electronic medical records according to claim 1, characterized in that: The weight acquisition module comprises: An average correlation degree acquisition unit, used to determine the average correlation degree corresponding to the decision tree according to the overall distribution level of the correlation degrees of all item features included in the decision tree; The reference weight acquisition unit is used to normalize the average correlation degree corresponding to the decision tree to obtain the reference weight of the decision tree, and the accumulated value of the reference weights of all the decision trees is equal to the value 1.

Images