Artificial intelligence-based medication recommendation method, device, equipment and storage medium

Abstract

The present invention discloses a medication recommendation method, device, equipment and storage medium based on artificial intelligence. The method includes: obtaining a medication recommendation request, the medication recommendation request includes a symptom type, current symptom data and user portrait data; obtaining a target medication recommendation model according to the target analysis result of the original medication recommendation model corresponding to the symptom type; using the target medication recommendation model to analyze and process the current symptom data to obtain the original recommended medication; performing feature analysis on the original recommended medication to obtain the feature analysis value corresponding to the feature to be analyzed; determining the recommendation tendency type based on the user portrait data, and obtaining the feature weight corresponding to the feature to be analyzed according to the recommendation tendency type; obtaining the recommendation evaluation value of the original recommended medication based on the feature analysis value and feature weight corresponding to the feature to be analyzed, and determining the original recommended medication with the largest recommendation evaluation value as the target recommended medication. The method can effectively improve the effectiveness and pertinence of intelligent medication recommendations.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based medication recommendation method, device, equipment and storage medium. Background Art

[0002] Clinical Decision Support System (CDSS) generally refers to any computer system that can provide support for clinical decision-making. This system makes full use of available and appropriate computer technology to improve and enhance decision-making efficiency through human-computer interaction for semi-structured or unstructured medical problems. CDSS is an important means to improve medical quality, that is, through the CDSS system to deeply analyze medical records, so as to make the most appropriate diagnosis and treatment decisions. Its fundamental purpose is to evaluate and improve medical quality, reduce medical errors, and thus control medical expenses.

[0003] The medication recommendation model currently integrated in the CDSS system is a model used to recommend medication by analyzing symptom data. It is mainly divided into two types: knowledge-driven medication recommendation model and data-driven medication recommendation model. The knowledge-driven medication recommendation model is a medication recommendation model based on various medical guidelines and expert consensus. The data-driven medication recommendation model is a medication recommendation model formed by integrating various artificial intelligence technologies such as machine learning and deep learning. The current evaluation of the quality of the medication recommendation model is mainly based on whether the recommended medication matches the actual clinical prescription medication. The medication recommendation model determined according to this standard is used to recommend medication. Due to the uneven level of clinical practice, the therapeutic efficacy of the actual clinical prescription medication is not good, which will reduce the recommendation accuracy of the medication recommendation model, that is, the effectiveness of the efficacy of the recommended medication obtained is not high, affecting the promotion of the medication recommendation model. Summary of the invention

[0004] The embodiments of the present invention provide a method, device, equipment and storage medium for medication recommendation based on artificial intelligence to solve the problem of low recommendation accuracy when the current medication recommendation model makes medication recommendations.

[0005] An artificial intelligence-based medication recommendation method, comprising:

[0006] Obtaining a medication recommendation request, wherein the medication recommendation request includes a symptom type, current symptom data, and user portrait data;

[0007] According to the target analysis results of the original medication recommendation model corresponding to the disease type, obtaining at least two target medication recommendation models with better target analysis results;

[0008] Using at least two of the target medication recommendation models to analyze and process the current symptom data to obtain at least two original recommended medications;

[0009] Performing feature analysis on the original recommended medication to obtain feature analysis values ​​corresponding to at least two features to be analyzed;

[0010] Determine a recommendation tendency type based on the user portrait data, and obtain feature weights corresponding to at least two features to be analyzed according to the recommendation tendency type;

[0011] Based on the feature analysis values ​​corresponding to at least two of the features to be analyzed and the feature weights, recommendation evaluation values ​​of at least two of the originally recommended medications are obtained, and the originally recommended medication with the largest recommendation evaluation value is determined as the target recommended medication.

[0012] A medication recommendation device based on artificial intelligence, comprising:

[0013] A recommendation request acquisition module, used to acquire a medication recommendation request, wherein the medication recommendation request includes a symptom type, current symptom data, and user portrait data;

[0014] A recommendation model determination module, configured to obtain at least two target medication recommendation models having better target analysis results according to the target analysis results of the original medication recommendation model corresponding to the disease type;

[0015] An original recommended medication acquisition module, used to analyze and process the current symptom data using at least two target medication recommendation models to acquire at least two original recommended medications;

[0016] A feature analysis value acquisition module, used to perform feature analysis on the original recommended medication and obtain feature analysis values ​​corresponding to at least two features to be analyzed;

[0017] A feature weight acquisition module, used to determine a recommendation tendency type based on the user portrait data, and acquire feature weights corresponding to at least two features to be analyzed according to the recommendation tendency type;

[0018] The target recommended medication acquisition module is used to obtain the recommendation evaluation values ​​of at least two of the original recommended medications based on the feature analysis values ​​corresponding to at least two of the features to be analyzed and the feature weights, and determine the original recommended medication with the largest recommendation evaluation value as the target recommended medication.

[0019] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned artificial intelligence-based medication recommendation method when executing the computer program.

[0020] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned artificial intelligence-based medication recommendation method.

[0021] The above-mentioned artificial intelligence-based medication recommendation method, device, equipment and storage medium obtain at least two target medication recommendation models with better target analysis results according to the target analysis results of the original medication recommendation model corresponding to the disease type, and then use the target medication recommendation model to analyze and process the current disease data to obtain the original recommended medication, which can ensure the effectiveness and pertinence of the original recommended medication; then perform feature analysis on the original recommended medication to determine the feature analysis value corresponding to the feature to be analyzed, determine the feature weight corresponding to the feature to be analyzed based on the recommendation tendency type determined by analyzing the user portrait data, and perform weighted calculations based on the feature analysis value corresponding to the feature to be analyzed and the corresponding feature weight to determine the recommendation evaluation value, so that the recommendation evaluation value can effectively reflect the effectiveness of the medication recommendation and the fit with the user portrait data, thereby improving the effectiveness and pertinence of the intelligent medication recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0023] Figure 1 is a schematic diagram of an application environment of a medication recommendation method based on artificial intelligence in one embodiment of the present invention;

[0024] Figure 2 is a flow chart of a medication recommendation method based on artificial intelligence in one embodiment of the present invention;

[0025] Figure 3 is another flow chart of a method for recommending medication based on artificial intelligence in one embodiment of the present invention;

[0026] Figure 4 is another flow chart of a method for recommending medication based on artificial intelligence in one embodiment of the present invention;

[0027] Figure 5 is another flow chart of a method for recommending medication based on artificial intelligence in one embodiment of the present invention;

[0028] Figure 6 is another flow chart of a method for recommending medication based on artificial intelligence in one embodiment of the present invention;

[0029] Figure 7is another flow chart of a method for recommending medication based on artificial intelligence in one embodiment of the present invention;

[0030] Figure 8 is another flow chart of a method for recommending medication based on artificial intelligence in one embodiment of the present invention;

[0031] Fig. 9 is a schematic diagram of a medication recommendation device based on artificial intelligence in one embodiment of the present invention;

[0032] Fig.10 is a schematic diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] The artificial intelligence-based medication recommendation method provided in the embodiment of the present invention can be applied as follows: Figure 1 Specifically, the artificial intelligence-based medication recommendation method is applied in a CDSS system, which includes: Figure 1 The client and server shown in the figure communicate with each other through the network to realize the intelligent analysis of the current disease data and user portrait data according to the medication recommendation request triggered by the client, so as to obtain the target recommended medication to ensure the accuracy, efficiency and pertinence of the recommended medication. Among them, the client, also known as the user end, refers to the program corresponding to the server that provides local services to customers. The client can be installed on but not limited to various personal computers, laptops, smart phones, tablets and portable wearable devices. The server can be implemented as an independent server or a server cluster consisting of multiple servers.

[0035] In one embodiment, if Figure 2 As shown, a method for recommending medication based on artificial intelligence is provided, and the method is applied in Figure 1 The server in the example is used as an example to illustrate the process, including the following steps:

[0036] S201: Obtain a medication recommendation request, which includes the disease type, current disease data, and user portrait data.

[0037] The medication recommendation request is a request for triggering the execution of the medication recommendation service. As an example, the server of the CDSS system can receive a medication recommendation request triggered by a user through a client, and obtain the disease type, current disease data, and user portrait data in the medication recommendation request.

[0038] The disease type is used to reflect the type of disease that the patient suffers from. As an example, the disease type includes but is not limited to diabetes, hypertension and chronic kidney disease. It can be understood that the determination of the disease type helps to determine a more accurate medication recommendation model for medication recommendation, thereby improving the accuracy of medication recommendation.

[0039] Among them, the current disease data is the data corresponding to the patient's current physical condition. The current disease data refers to the data corresponding to the detection indicators corresponding to the disease type detected by the patient in real time at the medical site. For example, for the disease type of diabetes, it is necessary to collect the measured data corresponding to indicators such as blood sugar, glycosylated hemoglobin, glycosylated serum protein and antibodies in real time.

[0040] Among them, user portrait data is portrait data used to reflect patient-related information. These user portrait data may include data that affects the diagnosis of the disease, such as age, gender and other data, and may also include data that affects medication recommendations. For example, when the patient's income is low, he or she may be more inclined to use low-cost medications.

[0041] S202: According to the target analysis results of the original medication recommendation model corresponding to the disease type, at least two target medication recommendation models with better target analysis results are obtained.

[0042] Among them, the original medication recommendation model is a medication recommendation model corresponding to the disease type that is pre-stored in the CDSS system, and is a model that can output medication recommendations based on disease data. That is, for each disease type, the CDSS system is connected to at least one original medication recommendation model corresponding to the disease type, so that the CDSS system can perform intelligent analysis on the current disease data corresponding to the disease type based on the pre-connected original medication recommendation model, and improve the pertinence and effectiveness of the recommended medication obtained by the analysis. It can be understood that different original medication recommendation models corresponding to the same disease type can be data obtained by using different neural network models to train clinical medical data (including clinical disease data and corresponding clinical medication data), so that each original medication recommendation model is based on clinical medical guidelines to avoid clinical errors in the recommended medication, and to use medical big data to realize personalized medication recommendations and learn the medication regimens of similar patients to ensure the effectiveness of medication recommendations.

[0043] Since different original medication recommendation models use different neural network models and have different training processes, different medication recommendation results may occur when the original medication recommendation models obtained through training perform intelligent medication recommendation for the same current disease data. Therefore, the CDSS system needs to conduct a comprehensive evaluation of each original medication recommendation model in advance, that is, conduct a comprehensive evaluation of the original medication recommendation model based on the clinical medical data of similar patients to obtain the target analysis results, and store the target analysis results in the CDSS system, so that the CDSS system can quickly query the target analysis results of the original medication recommendation model according to the disease type during the medication recommendation process, and sort all the original medication recommendation models in descending order according to the target analysis results, and select at least two original medication recommendation models with the target analysis results in front as the target medication recommendation models used for this medication recommendation. It can be understood that the number of target medication recommendation models can be set independently according to actual needs, and at least two original medication recommendation models in front of the target analysis results in descending order are determined as target medication recommendation models, so that when intelligent medication recommendations are performed based on the target medication recommendation model, it helps to ensure the effectiveness of its medication recommendation results.

[0044] S203: Analyze and process the current symptom data using at least two target medication recommendation models to obtain at least two original recommended medications.

[0045] The original recommended medication is the output result after the target medication recommendation model analyzes and processes the current symptom data. As an example, after obtaining at least two target medication recommendation models, the CDSS system uses at least two target medication recommendation models in parallel to analyze and process the current symptom data to obtain the original recommended medication output by the at least two target medication recommendation models.

[0046] S204: Perform feature analysis on the original recommended medication to obtain feature analysis values ​​corresponding to at least two features to be analyzed.

[0047] The features to be analyzed are pre-configured features that need to be analyzed and processed, and the features to be analyzed can be understood as features used to evaluate the quality of medication recommendations. In this example, the features to be analyzed include but are not limited to features such as medication cost, efficacy time, and side effects.

[0048] As an example, after obtaining at least two original recommended medications, the CDSS system uses a pre-set feature analysis strategy to perform feature analysis on the original recommended medications to obtain feature analysis values ​​corresponding to at least two features to be analyzed. In this example, the CDSS system uses a pre-set feature analysis strategy to perform feature analysis on the original recommended medications, which may include: first, based on the original recommended medication query system database or associated database, obtain the feature original data corresponding to the features to be analyzed in the original recommended medications, for example, obtain the actual cost corresponding to the feature to be analyzed, that is, the cost of medication required for a course of treatment. Next, the feature original data corresponding to all features to be analyzed are normalized to convert all feature original data corresponding to the features to be analyzed into dimensionless quantities to ensure the feasibility of subsequent recommended evaluation value calculations.

[0049] S205: Determine the recommendation tendency type based on the user portrait data, and obtain feature weights corresponding to at least two features to be analyzed according to the recommendation tendency type.

[0050] Among them, the recommendation tendency type is the medication recommendation type that reflects the patient's preference based on the patient's user portrait data. As an example, the medication recommendation type includes but is not limited to a low-cost recommendation type with a larger cost weight, a high-efficacy recommendation type with a larger efficacy time weight, and a low-risk recommendation type with a larger side effect weight. In this example, each medication recommendation type is pre-configured with feature weights corresponding to at least two features to be analyzed, and the corresponding feature weights are stored in the system database.

[0051] As an example, the CDSS system uses a pre-set recommendation tendency analysis strategy to perform intelligent analysis on user portrait data to determine the recommendation tendency type corresponding to the user portrait data; then, based on the recommendation tendency type, the recommendation tendency type is obtained to query the system database to obtain the pre-configured feature weights corresponding to each feature to be analyzed.

[0052] S206: Based on the feature analysis values ​​and feature weights corresponding to at least two features to be analyzed, obtain recommendation evaluation values ​​of at least two originally recommended medications, and determine the originally recommended medication with the largest recommendation evaluation value as the target recommended medication.

[0053] Among them, the target recommended medication is a recommended medication plan that is screened out from at least two original recommended medications and best matches the user portrait data.

[0054] In this embodiment, after the feature analysis values ​​and feature weights corresponding to at least two features to be analyzed are calculated, the CDSS system uses a weighted algorithm to perform weighted operations on the feature analysis values ​​and feature weights corresponding to at least two features to be analyzed, and obtains the recommended evaluation values ​​of at least two original recommended medications; then the at least two recommended evaluation values ​​are compared to determine the maximum value of the recommended evaluation values, and the original recommended medication corresponding to the maximum value of the recommended evaluation value is determined as the target recommended medication. The recommended evaluation value is a numerical value obtained after weighted operations are performed on the feature analysis values ​​and feature weights corresponding to at least two features to be analyzed, which can reflect the degree of fit with the patient's user portrait data. The higher the recommended evaluation value, the more the original recommended medication fits the patient's actual needs, thereby improving the pertinence and effectiveness of medication recommendations.

[0055] As an example, the CDSS system may adopt P = ∑S i W i The feature analysis values ​​and feature weights corresponding to at least two features to be analyzed are weighted to obtain the recommended evaluation values ​​of at least two originally recommended drugs. The process of obtaining the recommended evaluation values ​​is simple and convenient to calculate; wherein P is the recommended evaluation value, S i is the feature analysis value corresponding to the i-th feature to be analyzed, W i is the feature weight corresponding to the i-th feature to be analyzed.

[0056] As another example, after obtaining the target analysis results of each original medication recommendation model in step S202, the target analysis results can also be processed using a model weight conversion strategy to obtain the model weight corresponding to each target analysis result; then the CDSS system can use P = Q∑S i W i Perform a weighted operation on the feature analysis values ​​and feature weights corresponding to at least two features to be analyzed to obtain the recommended evaluation values ​​of at least two original recommended medications, where Q is the model weight corresponding to the original medication recommendation model, P is the recommended evaluation value, and S i is the feature analysis value corresponding to the i-th feature to be analyzed, W i is the feature weight corresponding to the i-th feature to be analyzed. It can be understood that the target analysis result of the original medication recommendation model is converted into a model weight, and then the model weight, feature analysis value and feature weight are used to determine the recommendation evaluation value, so that the recommendation evaluation value can effectively reflect the effectiveness of the medication recommendation and the fit with the user portrait data, thereby improving the effectiveness and pertinence of the intelligent medication recommendation.

[0057] In the artificial intelligence-based medication recommendation method provided in this embodiment, at least two target medication recommendation models with better target analysis results are obtained according to the target analysis results of the original medication recommendation model corresponding to the disease type, and then the target medication recommendation model is used to analyze and process the current disease data to obtain the original recommended medication, which can ensure the effectiveness and pertinence of the original recommended medication; then, feature analysis is performed on the original recommended medication to determine the feature analysis value corresponding to the feature to be analyzed, and the feature weight corresponding to the feature to be analyzed is determined according to the recommendation tendency type determined by analyzing the user portrait data, and a weighted operation is performed based on the feature analysis value corresponding to the feature to be analyzed and the corresponding feature weight to determine the recommendation evaluation value, so that the recommendation evaluation value can effectively reflect the effectiveness of the medication recommendation and the degree of fit with the user portrait data, thereby improving the effectiveness and pertinence of the intelligent medication recommendation.

[0058] In one embodiment, if Figure 3 As shown, before obtaining a medication recommendation request, the medication recommendation method based on artificial intelligence also includes the following steps:

[0059] S301: Acquire retrospective data corresponding to the disease type, where the retrospective data includes historical disease data and actual medication data.

[0060] The retrospective data refers to data used to evaluate the quality of the original medication recommendation model to determine the target analysis results. As an example, the retrospective data can be obtained from a system database of a hospital information system or a CDSS system, or it can be some public medical data sets, which are specific structured data.

[0061] In this example, the retrospective data is a visit record, that is, the data determined by the patient's one visit to the clinic, specifically including historical disease data and actual medication data. Historical disease data refers to the data related to the disease recorded when the patient visits the clinic, including symptoms and test results information. Actual medication data refers to the medication data given by the doctor during the visit, specifically including disease diagnosis results, prescription medications and treatment course time data. It can be understood that since the treatment effect of the disease symptoms corresponding to the historical disease data needs to be analyzed in combination with the actual medication data during the comprehensive evaluation of the original medication recommendation model, the retrospective data also includes short-term effect data and long-term effect data. Among them, short-term effect data refers to the data on the compliance of the test and inspection indicators in a short period of time after this visit. For example, the data on whether the glycated hemoglobin meets the standard about 3 months after treatment can be determined as short-term effect data. Long-term effect data refers to the data on whether there are complications or adverse events for a long period of time after this visit, and the proportion can be determined based on the subsequent outpatient or inpatient diagnosis results. It can be understood that the time limit of short-term effect data and long-term effect data can be determined based on historical experience.

[0062] As an example, the CDSS system can query the hospital information system or the system database of the CDSS system or crawl the hospital data set according to the disease type to obtain all historical medical records corresponding to the disease type; then verify all historical medical records to determine whether the historical medical records include necessary data such as historical disease data, actual medication data, short-term effect data and long-term effect data; if the historical medical records contain all necessary data, then the historical medical records are determined as retrospective data that can be used to evaluate the original medication recommendation model to ensure the feasibility of subsequent evaluation of the original medication recommendation model.

[0063] S302: Use the original medication recommendation model corresponding to the disease type to analyze and process the historical disease data, obtain the historical recommended medication data, match the historical recommended medication data with the actual medication data, obtain the matching results, and divide the retrospective data into a first data set that meets the model recommendation and a second data set that does not meet the model recommendation based on the matching results.

[0064] Since the original medication recommendation model is a model that can output medication recommendations based on symptom data, the CDSS system can use the original medication recommendation model to analyze and process historical symptom data to obtain historical recommended medication data. That is, the historical recommended medication data refers to the medication data output by analyzing and processing historical symptom data using the original medication recommendation model.

[0065] After obtaining the historical recommended medication data, the CDSS system can match the historical recommended medication data with the actual medication data to determine whether the historical recommended medication data matches the actual medication data, and divide the retrospective data into a first data set that meets the model recommendation and a second data set that does not meet the model recommendation according to the matching results. That is, if the historical recommended medication data matches the actual medication data, the retrospective data will be classified into the first data set that meets the model recommendation; if the historical recommended medication data does not match the actual medication data, the retrospective data will be classified into the second data set that does not meet the model recommendation.

[0066] S303: For the retrospective data in the first data set and the second data set, the short-term effectiveness index achievement rate and the long-term complication incidence rate are respectively counted, and significance verification is performed to obtain the effect analysis results.

[0067] Among them, the short-term effectiveness index compliance rate is used to evaluate the impact of model recommendations on whether the short-term effectiveness index meets the standard from the visit level or patient level based on retrospective data. The long-term complication rate is used to evaluate the impact of model recommendations on whether long-term complications occur from the visit level or patient level based on retrospective data. The visit level refers to a patient's medical record, which is mainly used to evaluate whether the short-term effectiveness index meets the standard, such as whether the short-term effectiveness index meets the standard within a short period of time after each visit. The patient level refers to all medical records for the same patient, which is mainly used to evaluate whether long-term complications occur. As an example, the recommended compliance rate of the actual medication data and the historical recommended medication data can be determined based on the multiple medical records of the same patient, and then the relationship between the recommended compliance rate at the patient level and the long-term complication rate can be further analyzed based on the recommended compliance rate. For example, if the same patient has a total of 10 visits, that is, 10 visit records are formed, of which 6 actual medication data are consistent with the historical recommended medication data, that is, the recommended compliance rate is 60%, and then the significance test is performed based on the compliance model recommendation rate, so as to obtain the effect analysis results reflecting the relationship between the compliance model recommendation rate and the long-term complication rate. A significance test is to make an assumption about the parameters or distribution form of the population (random variable) in advance, and then use sample information to determine whether this assumption (alternative hypothesis) is reasonable, that is, to determine whether there is a significant difference between the actual situation of the population and the original hypothesis.

[0068] As an example, the CDSS system counts the corresponding short-term effectiveness indicator achievement rates and long-term complication rates based on the retrospective data in the first data set and the second data set, respectively, so as to perform significance verification based on the statistical short-term effectiveness indicator achievement rates and long-term complication rates, thereby obtaining effect analysis results. The effect analysis results can effectively reflect the relationship between the short-term effectiveness indicator achievement rates and the long-term complication rates and the model compliance rate, so as to conduct subsequent comprehensive evaluation.

[0069] S304: If the effect analysis result meets the preset analysis conditions, a difference analysis is performed on the retrospective data in the first data set and the second data set to obtain the target short-term confounding factors and the target long-term confounding factors.

[0070] The preset analysis condition is a pre-set condition for subsequent difference analysis. As an example, when the effect analysis result is a P-value obtained by significance verification, the preset analysis result can be set to be less than a preset value, such as less than 0.05.

[0071] Since the differences in the short-term effectiveness index achievement rate and long-term complication rate between the retrospective data in the first data set and the second data set may be caused by treatment factors or non-treatment factors, and treatment factors are related to medication, while non-treatment factors are not related to medication, in order to avoid the impact of non-treatment factors on the effectiveness of the medication recommended by the model, it is necessary to perform a difference analysis on the retrospective data in the first data set and the second data set to obtain the target short-term confounding factors and target long-term confounding factors. Confounding factors can be understood as non-treatment factors; target short-term confounding factors refer to the non-treatment factors that have the greatest or greater impact on the short-term effectiveness index achievement rate; target long-term confounding factors refer to the non-treatment factors that have the greatest or greater impact on the long-term complication rate.

[0072] For example, the retrospective data include N confounding factors such as F1, F2, F3...Fn. When performing differential analysis on the retrospective data in the first data set and the second data set, for any confounding factor, its corresponding short-term effectiveness indicator achievement rate and long-term complication rate are respectively counted to obtain the effect analysis result corresponding to the confounding factor; then, based on the effect analysis results, the confounding factors with the greatest or greater impact on the short-term effectiveness indicator achievement rate are determined as target short-term confounding factors, and the confounding factors with the greatest or greater impact on the long-term complication rate are determined as target short-term confounding factors.

[0073] S305: Based on the target short-term confounding factors and the target short-term confounding factors, a tendency analysis is performed on the retrospective data in the first data set and the second data set to obtain and store the target analysis results of the original medication recommendation model.

[0074] In this example, based on the target short-term confounders and target long-term confounders determined by analysis, a tendency analysis is performed on the retrospective data in the first data set and the second data set to weaken the impact of confounders on the achievement rate of short-term effectiveness indicators and the incidence rate of long-term complications, so that the obtained target analysis results are reasonable. In this example, after the CDSS system obtains the target analysis results of the original medication recommendation model, it needs to store the target analysis results in the system database, so that after the CDSS system obtains the medication recommendation request, it can call the target analysis results of the original medication recommendation model and quickly determine at least two target medication recommendation models with better target analysis results, thereby ensuring the effectiveness of selecting a target medication recommendation model that has been clinically verified and learns the medication regimens of similar patients.

[0075] In one embodiment, if Figure 4 As shown, the matching process of the historical recommended medication data and the actual medication data in step S302 to obtain the matching result specifically includes the following steps:

[0076] S401: Acquire at least one recommended medication category based on historical recommended medication data, and acquire at least one actual medication category based on actual medication data.

[0077] S402: If all recommended medication categories match all actual medication categories, the matching result is consistent with the model recommendation.

[0078] S403: If there is at least one recommended medication category that does not match the actual medication category, the matching result is not in line with the model recommendation.

[0079] In this example, the recommended medication category can be determined based on the historical recommended medication data, and the actual medication category can be determined based on the actual medication data, and a comparison can be made as to whether all recommended medication categories and all actual medication categories match; if all recommended medication categories match all actual medication categories, that is, the combination of the recommended medication categories matches the combination of the actual medication categories, then the matching result is consistent with the model recommendation; if there is at least one recommended medication category that matches the actual medication category, that is, the combination of the recommended medication categories does not match the combination of the actual medication categories, then the matching result is inconsistent with the model recommendation.

[0080] For example, diabetes medications can be divided into biguanides, sulfonylureas and other medication categories. If the patient's actual medication includes metformin, it is considered that biguanides are used. It is understandable that the actual medication data can be defined as a combination of multiple medication categories, such as biguanides + sulfonylureas, that is, drugs of a single medication category can be used or drugs of multiple medication categories can be used at the same time. The matching of historical recommended medication data and doctors' actual medication data means that they are completely consistent in the combination of medication categories. If there is a difference in the number of medication categories or the name of the medication categories, it is considered to be mismatched. It is understandable that because different drugs have different efficacy under the same medication category, the dosage of the drugs is inconsistent. Therefore, there is no need to consider whether the dosage of the drugs matches. Therefore, it is possible to determine whether the historical recommended medication data and the actual medication data match based on whether the medication categories match, which helps to simplify the calculation amount of the matching process and improve processing efficiency.

[0081] In one embodiment, the retrospective data also includes short-term effect data and long-term effect data. Figure 5 As shown, in step S303, the short-term effectiveness index achievement rate and the long-term complication incidence rate are respectively counted for the retrospective data in the first data set and the second data set, and significance verification is performed to obtain the effect analysis results, which specifically includes the following steps:

[0082] S501: Based on the retrospective data in the first data set, obtain the first short-term effectiveness indicator achievement rate and the first long-term complication incidence rate.

[0083] S502: Based on the retrospective data in the second data set, obtain the second short-term effectiveness indicator achievement rate and the second long-term complication incidence rate.

[0084] S503: Perform significance verification on the first short-term effectiveness indicator achievement rate, the first long-term complication rate, the second short-term effectiveness indicator achievement rate, and the second long-term complication rate to obtain effect analysis results.

[0085] Among them, the short-term effectiveness index compliance rate refers to the ratio of the number of samples that meet the short-term effectiveness index to the total number of all samples. The long-term complication incidence rate refers to the ratio of the number of samples that develop long-term complications to the total number of all samples.

[0086] As an example, suppose the number of retrospective data corresponding to the disease type obtained by the CDSS system is A. The CDSS system can divide the first data set and the second data set according to the matching results of the historical recommended medication data and the actual medication data. Suppose the number of retrospective data in the first data set is A1, and the number of retrospective data in the second data set is A2.

[0087] Accordingly, the CDSS system will judge whether the short-term effect data in the retrospective data reaches the standard range corresponding to the short-term effectiveness indicator. If the short-term effect data reaches the standard range corresponding to the short-term effectiveness indicator, the short-term effectiveness indicator is deemed to have met the standard, and the number of samples that meet the short-term effectiveness indicator is updated, that is, the number of samples that meet the short-term effectiveness indicator is increased by 1 until all retrospective data are analyzed, and the number of samples that meet the short-term effectiveness indicator is B, among which the number of samples that meet the short-term effectiveness indicator in the first data set is B1, and the number of samples that meet the short-term effectiveness indicator in the second data set is B2.

[0088] Accordingly, the CDSS system will determine whether the criteria for long-term complications are met based on the long-term effect data in the retrospective data. If the long-term effect data meet the criteria for long-term complications, a long-term complication will be deemed to have occurred, and the number of samples of long-term complications will be updated, that is, the number of samples of long-term complications will be increased by 1 until all retrospective data are analyzed, and the number of samples of long-term complications is C, where the number of samples of long-term complications in the first data set is C1, and the number of samples of long-term complications in the second data set is C2.

[0089] In this example, the short-term effectiveness index achievement rate and the long-term complication rate are respectively counted for the retrospective data in the first data set and the second data set, and a significance check is performed, specifically including: (1) Based on the retrospective data in the first data set, the CDSS system obtains the first short-term effectiveness index achievement rate and the first long-term complication rate. Let the first short-term effectiveness index achievement rate be Q1, then Q1 = B1 / A1; let the first long-term complication rate be P1, then P1 = C1 / A1. (2) Based on the retrospective data in the second data set, the CDSS system obtains the second short-term effectiveness index achievement rate and the second long-term complication rate. Let the second short-term effectiveness index achievement rate be Q2, then Q2 = B2 / A2; let the second long-term complication rate be P2, then P2 = C2 / A2. (3) The CDSS system uses the chi-square test method to perform a significance check on the first short-term effectiveness index achievement rate Q1, the first long-term complication rate P1, the second short-term effectiveness index achievement rate Q2, and the second long-term complication rate P2 to obtain the effect analysis results.

[0090] As shown in Table 1 below, for all retrospective data on the achievement of short-term effectiveness indicators, if they meet the model recommendation, they are marked as "yes"; if they do not meet the model recommendation, they are marked as "no", and the number of samples B1 and B2 in the first data set and the second data set are counted, and then the first short-term effectiveness indicator achievement rate Q1 and the second short-term effectiveness indicator achievement rate Q2 are calculated based on B1 and B2, and then the difference between Q1 and Q2 is used to calculate P-value1 through significance test. Correspondingly, for the retrospective data on the occurrence of long-term complications, if they meet the model recommendation, they are marked as "yes"; if they do not meet the model recommendation, they are marked as "no", and the number of samples C1 and C2 in the first data set and the second data set are counted, and then the first long-term complication rate P1 and the second long-term complication rate P2 are calculated based on C1 and C2, and then the difference between P1 and P2 is used to calculate P-value2 through significance test. In this example, compliance with the model recommendation is used as the exposure variable, whether the short-term effectiveness indicators are met and whether long-term complications occur are used as outcome variables. The P-value between the exposure variable and the outcome variable is calculated using the chi-square test method. If the P-value < 0.05, it means that there is significance between the exposure variable and the outcome variable. The effect analysis results are obtained to achieve the purpose of analyzing the relationship between whether the model recommendation is met and the rate of achievement of short-term effectiveness indicators and the incidence of long-term complications.

[0091]

[0092] In one embodiment, if Figure 6As shown, step S304, i.e., performing difference analysis on the retrospective data in the first data set and the second data set to obtain target short-term confounding factors and target long-term confounding factors, specifically includes the following steps:

[0093] S601: Determine the confounding factors to be analyzed.

[0094] Among them, the confounding factors to be analyzed refer to the confounding factors that need to be analyzed and processed this time. For example, if the retrospective data includes N confounding factors such as F1, F2, F3...Fn, one confounding factor can be randomly selected as the confounding factor to be analyzed each time. For example, for the glycated target rate of diabetes, the initial glycated hemoglobin can be determined as its confounding factor to be analyzed.

[0095] S602: Based on the confounding factors to be analyzed, screen out a first analysis subset and a second analysis subset corresponding to the confounding factors to be analyzed from the retrospective data in the first data set and the second data set.

[0096] In this example, the CDSS system can divide the first data set into multiple first analysis subsets corresponding to the classification criteria according to the pre-set classification criteria for the confounding factors to be analyzed, and divide the second data set into multiple second analysis subsets corresponding to the classification criteria, and then divide the retrospective data into corresponding first analysis subsets and second analysis subsets based on whether the specific values ​​corresponding to the confounding factors in the retrospective data meet the corresponding classification criteria.

[0097] S603: For the retrospective data in the first analysis subset and the second analysis subset, the short-term effectiveness index achievement rate and the long-term complication incidence rate are respectively counted, and significance verification is performed, and the obtained effect analysis results are determined as the confounding analysis results corresponding to the confounding factors to be analyzed.

[0098] It can be understood that the actual process of step S603 is basically the same as that of S303. The detailed implementation scheme can refer to steps S501-S503. The difference lies in the different data sets. To avoid repetition, they are not described here one by one.

[0099] S604: Obtain target short-term confounding factors and target long-term confounding factors according to the confounding analysis results corresponding to the confounding factors to be analyzed.

[0100] In this example, after the short-term effectiveness index achievement rate and long-term complication rate are statistically analyzed and significance verified for the retrospective data in the first analysis subset and the second analysis subset, the corresponding effect analysis results are determined as the confounding analysis results corresponding to the confounding factors to be analyzed, and then according to the confounding analysis results, the confounding factors with the greatest or greater impact on the short-term effectiveness index achievement rate are determined as the target short-term confounding factors, and the confounding factors with the greatest or greater impact on the long-term complication rate are determined as the target short-term confounding factors. For example, for the glycated target rate, the initial glycated hemoglobin is the largest confounding factor, so it is determined as the target short-term confounding factor; for the complication rate, its initial concurrent risk is the largest confounding factor, so it is determined as the target long-term confounding factor.

[0101] In one embodiment, if Figure 7 As shown, step S305, i.e., based on the target short-term confounding factor and the target short-term confounding factor, performing a tendency analysis on the retrospective data in the first data set and the second data set, obtaining and storing the target analysis result of the original medication recommendation model, specifically includes the following steps:

[0102] S701: stratify the first data set and the second data set based on the target short-term confounding factor and the target short-term confounding factor to obtain a stratified data set.

[0103] S702: Perform a tendency analysis on the retrospective data in the stratified data set to obtain a tendency result corresponding to each stratified data set.

[0104] S703: Based on the tendency results corresponding to the stratified data set, the target analysis results corresponding to the original medication recommendation model are obtained and stored.

[0105] Among them, the stratified data set is a stratified data set formed by dividing the first data set and the second data set according to the target short-term confounding factor and the target long-term confounding factor. As an example, when the initial glycated hemoglobin is the target short-term confounding factor, the initial glycated hemoglobin can be divided into three stratified data sets of initial compliance, initial excess and severe excess, and the compliance rate of the short-term effectiveness index in each stratified data set that meets the model recommendation and does not meet the model recommendation is compared, that is, as a treatment group (treat group) and a control group (i.e., control group) are compared, and other confounding factors are controlled by a propensity score matching method to obtain the propensity results in each stratified data set. As another example, when the initial concurrent risk is the target long-term confounding factor, the initial concurrent risk can be divided into three stratified data sets of low risk, medium risk and high risk, and the incidence of long-term complications in each stratified data set that meets the model recommendation and does not meet the model recommendation is compared, that is, as a treatment group (treat group) and a control group (i.e., control group) are compared, and other confounding factors are controlled by a propensity score matching method to obtain the propensity results in each stratified data set.

[0106] In this example, after obtaining the tendency results in each stratified data set, the CDSS system compares the tendency results with the pre-set qualification assessment threshold for judging whether the results are qualified, thereby determining whether the corresponding stratified data set is qualified; then, the ratio of the qualified number to the total number corresponding to the stratified data set is counted to obtain the target analysis results, so as to achieve an objective analysis of the efficacy effectiveness of the original medication recommendation model based on the objectively existing short-term effect data and long-term effect data in the retrospective data, thereby ensuring the objectivity of the target analysis results.

[0107] In one embodiment, if Figure 8 As shown, step S205, i.e. determining the recommendation tendency type based on the user portrait data, specifically includes the following steps:

[0108] S801: Determine whether the user portrait data contains existing tendency types.

[0109] S802: If the user portrait data includes an existing tendency type, the existing tendency type is determined as a recommended tendency type.

[0110] S803: If the user portrait data does not contain the existing tendency type, similar groups of people are determined based on the user portrait data, and the common tendency type corresponding to the similar groups of people is determined as the recommended tendency type; or, related web page data is obtained based on the user portrait data, and the feature tags of the related web page data are counted to obtain the recommended tendency type.

[0111] The existing tendency type refers to the tendency type clearly indicated in the user portrait data. For example, the CDSS system input interface is configured with existing tendency types for users to choose from, so that users can configure the corresponding tendency type through the input interface.

[0112] As an example, the CDSS system uses a character matching algorithm or other algorithms to analyze the user portrait data to determine whether the user portrait data includes an existing tendency type. If the user portrait data contains an existing tendency type, the existing tendency type is directly determined as the recommended tendency type. Based on the recommended tendency type, the feature weights corresponding to each feature to be analyzed in the original recommended medication are determined, so that the target recommended medication determined based on the final recommended evaluation value is more in line with the user portrait data, which helps to improve the pertinence and effectiveness of the target recommended medication.

[0113] As an example, the CDSS system uses a character matching algorithm or other algorithm to analyze the user portrait data to determine whether the user portrait data includes an existing tendency type. If the user portrait data does not include an existing tendency type, similar populations are determined based on the user portrait data, and the common tendency type corresponding to the similar population is determined as the recommended tendency type. Among them, the similar population is the population that is closest to or most similar to the patient's user portrait data, and the similar population specifically refers to the population that is closest to or most similar to the user portrait data corresponding to the disease corresponding to the symptom type. The similar population includes a plurality of similar users that are closest to or most similar to the patient's user portrait data, and each similar user corresponds to a medication tendency type, which refers to a tendency type that the similar user has already determined. The common tendency type corresponding to the similar population refers to the medication tendency type with the largest proportion determined by statistics of the medication tendency types of all similar users in all similar populations.

[0114] For example, user portrait data includes but is not limited to gender, age, address, residence, occupation, consumption habits and exercise data. The CDSS system determines similar populations based on user portrait data, and the process of determining the common tendency type corresponding to the similar population as the recommended tendency type includes: (1) using a distance algorithm to calculate the distance between the patient's user portrait data and the user portrait data of any existing user stored in the system database, and obtaining the similarity distance corresponding to the two user portrait data. (2) Existing users whose similarity distance reaches a preset distance threshold are determined as similar users, and similar populations are formed based on all similar users. (3) Then, a statistical analysis is performed on the medication tendency types of all similar users in the similar population, and the medication tendency type with the largest proportion is determined to be the common tendency type, and the common tendency type is determined as the recommended tendency type for the patient. The distance algorithm is not limited to the Euclidean distance algorithm. It can be understood that since the user portrait data of the similar population is closest to or most similar to the user portrait data corresponding to the patient, their tendency types are also relatively similar, so the common tendency type of the similar population can be determined as the recommended tendency type for the patient.

[0115] As an example, the CDSS system uses a character matching algorithm or other algorithm to analyze the user portrait data to determine whether the user portrait data includes an existing tendency type. If the user portrait data does not include an existing tendency type, then the associated web page data is obtained based on the user portrait data, and the feature labels of the associated web page data are counted to obtain the recommended tendency type. Among them, the associated web page data refers to the content data in the associated web page related to the disease treatment corresponding to the symptom type. The feature label of the associated web page data refers to the label related to the feature to be analyzed that is pre-configured for the content data in each associated web page, for example, labels such as low cost, short efficacy time, and few side effects.

[0116] For example, based on user portrait data, associated web page data is obtained, and the feature tags of the associated web page data are counted. The process of obtaining the recommended tendency type includes: (1) Based on the ID number and / or mobile phone number in the user portrait data, all historical web page data visited by the patient corresponding to the ID number and / or mobile phone number are obtained. (2) The historical web page data related to the disease type in the user portrait data in all historical web page data are determined as associated web page data. (3) The number of feature tags in all associated web page data is counted, and the recommended tendency type is determined based on the number. It can be understood that the feature tags in each associated web page data can be obtained using the Jieba word segmentation tool and the TF-IDF algorithm, that is, the Jieba word segmentation tool can be used to scan, segment and tag the text information in the associated web page data to obtain the word segmentation results; then the TF-IDF algorithm can be used to extract keywords from the word segmentation results to obtain the feature tags corresponding to the associated web page data.

[0117] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0118] In one embodiment, a medication recommendation device based on artificial intelligence is provided, and the medication recommendation device based on artificial intelligence corresponds to the medication recommendation method based on artificial intelligence in the above embodiment. Fig. 9 As shown, the AI-based medication recommendation device includes a recommendation request acquisition module 901, a recommendation model determination module 902, an original recommended medication acquisition module 903, a feature analysis value acquisition module 904, a feature weight acquisition module 905, and a target recommended medication acquisition module 906. Each functional module is described in detail as follows:

[0119] The recommendation request acquisition module 901 is used to obtain a medication recommendation request, which includes a disease type, current disease data, and user portrait data.

[0120] The recommendation model determination module 902 is used to obtain at least two target medication recommendation models with better target analysis results based on the target analysis results of the original medication recommendation model corresponding to the disease type.

[0121] The original recommended medication acquisition module 903 is used to analyze and process the current symptom data using at least two target medication recommendation models to obtain at least two original recommended medications.

[0122] The feature analysis value acquisition module 904 is used to perform feature analysis on the original recommended medication and obtain feature analysis values ​​corresponding to at least two features to be analyzed.

[0123] The feature weight acquisition module 905 is used to determine the recommendation tendency type based on the user portrait data, and obtain the feature weights corresponding to at least two features to be analyzed according to the recommendation tendency type.

[0124] The target recommended medication acquisition module 906 is used to obtain the recommendation evaluation values ​​of at least two original recommended medications based on the feature analysis values ​​and feature weights corresponding to at least two features to be analyzed, and determine the original recommended medication with the largest recommendation evaluation value as the target recommended medication.

[0125] Preferably, the artificial intelligence-based medication recommendation device also includes a retrospective data acquisition module, a data partitioning module, an effect analysis result acquisition module, a confounding factor determination module and a target analysis result acquisition module.

[0126] The retrospective data acquisition module is used to acquire retrospective data corresponding to the disease type, and the retrospective data includes historical disease data and actual medication data.

[0127] The data partitioning module is used to analyze and process historical disease data using the original medication recommendation model corresponding to the disease type, obtain historical recommended medication data, match the historical recommended medication data with the actual medication data, obtain matching results, and based on the matching results, divide the retrospective data into a first data set that meets the model recommendations and a second data set that does not meet the model recommendations.

[0128] The effect analysis result acquisition module is used to count the short-term effectiveness index achievement rate and the long-term complication incidence rate for the retrospective data in the first data set and the second data set, and perform significance verification to obtain the effect analysis results.

[0129] The confounding factor determination module is used to perform a difference analysis on the retrospective data in the first data set and the second data set to obtain target short-term confounding factors and target long-term confounding factors if the effect analysis results meet the preset analysis conditions.

[0130] The target analysis result acquisition module is used to perform a tendency analysis on the retrospective data in the first data set and the second data set based on the target short-term confounders and the target short-term confounders, and to acquire and store the target analysis results of the original medication recommendation model.

[0131] Preferably, the data division module includes a medication category acquisition unit, a first matching result acquisition unit and a second matching result acquisition unit.

[0132] The medication category acquisition unit is used to acquire at least one recommended medication category based on historical recommended medication data, and to acquire at least one actual medication category based on actual medication data.

[0133] The first matching result obtaining unit is used to determine that if all recommended medication categories match all actual medication categories, the matching result is consistent with the model recommendation.

[0134] The second matching result obtaining unit is used for determining that the matching result does not conform to the model recommendation if at least one recommended medication category does not match the actual medication category.

[0135] Preferably, the retrospective data also includes short-term effect data and long-term effect data.

[0136] The effect analysis result acquisition module includes a first indicator acquisition unit, a second indicator acquisition unit and a significance verification unit.

[0137] The first indicator acquisition unit is used to acquire a first short-term effectiveness indicator compliance rate and a first long-term complication incidence rate based on the retrospective data in the first data set.

[0138] The second indicator acquisition unit is used to acquire a second short-term effectiveness indicator compliance rate and a second long-term complication incidence rate based on the retrospective data in the second data set.

[0139] The significance verification unit is used to perform significance verification on the first short-term effectiveness indicator achievement rate, the first long-term complication incidence rate, the second short-term effectiveness indicator achievement rate and the second long-term complication incidence rate to obtain effect analysis results.

[0140] Preferably, the confounding factor determination module includes a to-be-analyzed factor determination unit, an analysis subset division unit, a confounding analysis result acquisition unit and a target confounding factor acquisition unit.

[0141] The unit for determining factors to be analyzed is used to determine the confounding factors to be analyzed.

[0142] The analysis subset division unit is used to screen out the first analysis subset and the second analysis subset corresponding to the confounding factors to be analyzed from the retrospective data in the first data set and the second data set based on the confounding factors to be analyzed.

[0143] The confounding analysis result acquisition unit is used to count the short-term effectiveness indicator achievement rate and the long-term complication incidence rate for the retrospective data in the first analysis subset and the second analysis subset, and perform significance verification, and determine the obtained effect analysis results as the confounding analysis results corresponding to the confounding factors to be analyzed.

[0144] The target confounding factor acquisition unit is used to acquire the target short-term confounding factor and the target long-term confounding factor according to the confounding analysis results corresponding to the confounding factor to be analyzed.

[0145] Preferably, the target analysis result acquisition module includes a hierarchical data set acquisition unit, a tendency result acquisition unit and a target analysis result acquisition unit.

[0146] The stratified data set acquisition unit is used to stratify the first data set and the second data set based on the target short-term confounding factor and the target short-term confounding factor to acquire the stratified data set.

[0147] The tendency result acquisition unit is used to perform tendency analysis on the retrospective data in the stratified data set to obtain the tendency result corresponding to each stratified data set.

[0148] The target analysis result acquisition unit is used to acquire and store the target analysis results corresponding to the original medication recommendation model based on the tendency results corresponding to the stratified data set.

[0149] Preferably, the feature weight acquisition module 905 includes an existing tendency judgment unit, a first tendency type determination unit and a second tendency type determination unit.

[0150] The existing tendency judgment unit is used to judge whether the user portrait data contains an existing tendency type.

[0151] The first tendency type determination unit is used to determine the existing tendency type as a recommended tendency type if the user portrait data contains an existing tendency type.

[0152] The second tendency type determination unit is used to determine similar groups of people based on the user portrait data if the user portrait data does not contain the existing tendency type, and determine the common tendency type corresponding to the similar groups of people as the recommended tendency type; or, obtain related web page data based on the user portrait data, count the feature tags of the related web page data, and obtain the recommended tendency type.

[0153] For the specific limitations of the AI-based medication recommendation device, please refer to the limitations of the AI-based medication recommendation method above, which will not be repeated here. Each module in the above-mentioned AI-based medication recommendation device can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0154] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.10 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data used or generated in the process of executing the artificial intelligence-based medication recommendation method. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for recommending medication based on artificial intelligence is implemented.

[0155] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for recommending medication based on artificial intelligence in the above embodiment is implemented, for example Figure 2 S201-S206 as shown, or Figures 3 to 8 Alternatively, when the processor executes the computer program, the functions of each module / unit in the embodiment of the drug recommendation device based on artificial intelligence are realized, for example Fig. 9The functions of the modules / units shown are not described here in detail to avoid repetition.

[0156] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for recommending medication based on artificial intelligence in the above embodiment is implemented, for example Figure 2 S201-S206 as shown, or Figures 3 to 8 Alternatively, when the computer program is executed by the processor, the functions of each module / unit in the embodiment of the above-mentioned artificial intelligence-based medication recommendation device are realized, for example Fig. 9 The functions of the modules / units shown are not described here in detail to avoid repetition.

[0157] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0158] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0159] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and 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 medication recommendation method based on artificial intelligence, characterized in that: include: Acquiring retrospective data corresponding to the disease type, wherein the retrospective data includes historical disease data and actual medication data; Using an original medication recommendation model corresponding to the disease type to analyze and process the historical disease data, obtain historical recommended medication data, match the historical recommended medication data with the actual medication data to obtain a matching result, and divide the retrospective data into a first data set that meets the model recommendation and a second data set that does not meet the model recommendation based on the matching result; For the retrospective data in the first data set and the second data set, statistically analyze the short-term effectiveness index achievement rate and the long-term complication incidence rate, and perform significance verification to obtain effect analysis results; If the effect analysis result meets the preset analysis conditions, a difference analysis is performed on the retrospective data in the first data set and the second data set to obtain target short-term confounding factors and target long-term confounding factors. The target short-term confounding factors refer to the non-therapeutic factors with the greatest or the largest impact on the achievement rate of short-term effectiveness indicators; the target long-term confounding factors refer to the non-therapeutic factors with the greatest or the largest impact on the incidence of long-term complications; Based on the target short-term confounding factor and the target short-term confounding factor, performing a tendency analysis on the retrospective data in the first data set and the second data set, and obtaining and storing a target analysis result of the original medication recommendation model; Obtaining a medication recommendation request, wherein the medication recommendation request includes a symptom type, current symptom data, and user portrait data; According to the target analysis results of the original medication recommendation model corresponding to the disease type, obtaining at least two target medication recommendation models with better target analysis results; Using at least two of the target medication recommendation models to analyze and process the current symptom data to obtain at least two original recommended medications; Performing feature analysis on the original recommended medication to obtain feature analysis values ​​corresponding to at least two features to be analyzed; Determine a recommendation tendency type based on the user portrait data, and obtain feature weights corresponding to at least two features to be analyzed according to the recommendation tendency type; Based on the feature analysis values ​​corresponding to at least two of the features to be analyzed and the feature weights, recommendation evaluation values ​​of at least two of the originally recommended medications are obtained, and the originally recommended medication with the largest recommendation evaluation value is determined as the target recommended medication.

2. The method for recommending medication based on artificial intelligence according to claim 1, characterized in that: The matching process of the historical recommended medication data with the actual medication data to obtain a matching result includes: Acquire at least one recommended medication category based on the historical recommended medication data, and acquire at least one actual medication category based on the actual medication data; If all the recommended medication categories match all the actual medication categories, then the matching result is in line with the model recommendation; If there is at least one mismatch between the recommended medication category and the actual medication category, the matching result is not in compliance with the model recommendation.

3. The method for recommending medication based on artificial intelligence according to claim 1, characterized in that: The retrospective data also includes short-term effect data and long-term effect data; The retrospective data in the first data set and the second data set are respectively statistically analyzed for the short-term effectiveness index achievement rate and the long-term complication incidence rate, and significance verification is performed to obtain effect analysis results, including: Based on the retrospective data in the first data set, obtaining the first short-term effectiveness indicator achievement rate and the first long-term complication incidence rate; Based on the retrospective data in the second data set, obtaining the second short-term effectiveness indicator achievement rate and the second long-term complication incidence rate; A significance check is performed on the first short-term effectiveness indicator achievement rate, the first long-term complication incidence rate, the second short-term effectiveness indicator achievement rate and the second long-term complication incidence rate to obtain effect analysis results.

4. The method for recommending medication based on artificial intelligence according to claim 1, characterized in that: The performing difference analysis on the retrospective data in the first data set and the second data set to obtain target short-term confounding factors and target long-term confounding factors includes: Identify confounding factors to be analyzed; Based on the confounding factors to be analyzed, screening out a first analysis subset and a second analysis subset corresponding to the confounding factors to be analyzed from the retrospective data in the first data set and the second data set; For the retrospective data in the first analysis subset and the second analysis subset, statistically analyze the short-term effectiveness index achievement rate and the long-term complication incidence rate, respectively, and perform significance verification, and determine the obtained effect analysis results as the confounding analysis results corresponding to the confounding factors to be analyzed; According to the confounding analysis results corresponding to the confounding factors to be analyzed, target short-term confounding factors and target long-term confounding factors are obtained.

5. The method for recommending medication based on artificial intelligence according to claim 1, characterized in that: The step of performing a tendency analysis on the retrospective data in the first data set and the second data set based on the target short-term confounding factor and the target short-term confounding factor, and acquiring and storing the target analysis result of the original medication recommendation model includes: stratifying the first data set and the second data set based on the target short-term confounding factor and the target short-term confounding factor to obtain a stratified data set; Performing a tendency analysis on the retrospective data in the stratified data set to obtain a tendency result corresponding to each stratified data set; Based on the tendency results corresponding to the stratified data set, the target analysis results corresponding to the original medication recommendation model are obtained and stored.

6. The method for recommending medication based on artificial intelligence according to claim 1, characterized in that: The determining of the recommendation tendency type based on the user portrait data includes: Determine whether the user portrait data contains an existing tendency type; If the user portrait data includes an existing tendency type, determining the existing tendency type as a recommended tendency type; If the user portrait data does not contain the existing tendency type, similar groups of people are determined based on the user portrait data, and the common tendency type corresponding to the similar groups of people is determined as the recommended tendency type; or, related web page data is obtained based on the user portrait data, and the feature tags of the related web page data are counted to obtain the recommended tendency type.

7. An artificial intelligence-based medication recommendation device, characterized in that: include: A retrospective data acquisition module, used to acquire retrospective data corresponding to the disease type, wherein the retrospective data includes historical disease data and actual medication data; a data partitioning module, configured to analyze and process the historical disease data using an original medication recommendation model corresponding to the disease type, obtain historical recommended medication data, match the historical recommended medication data with the actual medication data, obtain matching results, and partition the retrospective data into a first data set that meets the model recommendation and a second data set that does not meet the model recommendation based on the matching results; An effect analysis result acquisition module is used to count the short-term effectiveness index achievement rate and the long-term complication incidence rate for the retrospective data in the first data set and the second data set, and perform significance verification to obtain effect analysis results; A confounding factor determination module, for performing a difference analysis on the retrospective data in the first data set and the second data set if the effect analysis result meets the preset analysis conditions, to obtain target short-term confounding factors and target long-term confounding factors, wherein the target short-term confounding factors refer to the non-therapeutic factors with the greatest or the largest influence on the achievement rate of short-term effectiveness indicators; and the target long-term confounding factors refer to the non-therapeutic factors with the greatest or the largest influence on the incidence rate of long-term complications; a target analysis result acquisition module, configured to perform a tendency analysis on the retrospective data in the first data set and the second data set based on the target short-term confounding factor and the target short-term confounding factor, and to acquire and store the target analysis result of the original medication recommendation model; A recommendation request acquisition module, used to acquire a medication recommendation request, wherein the medication recommendation request includes a symptom type, current symptom data, and user portrait data; A recommendation model determination module, configured to obtain at least two target medication recommendation models having better target analysis results according to the target analysis results of the original medication recommendation model corresponding to the disease type; An original recommended medication acquisition module, used to analyze and process the current symptom data using at least two target medication recommendation models to acquire at least two original recommended medications; A feature analysis value acquisition module, used to perform feature analysis on the original recommended medication and obtain feature analysis values ​​corresponding to at least two features to be analyzed; A feature weight acquisition module, used to determine a recommendation tendency type based on the user portrait data, and acquire feature weights corresponding to at least two features to be analyzed according to the recommendation tendency type; The target recommended medication acquisition module is used to obtain the recommendation evaluation values ​​of at least two of the original recommended medications based on the feature analysis values ​​corresponding to at least two of the features to be analyzed and the feature weights, and determine the original recommended medication with the largest recommendation evaluation value as the target recommended medication.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the artificial intelligence-based medication recommendation method as described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the artificial intelligence-based medication recommendation method as described in any one of claims 1 to 6 is implemented.

Images