An implicit rule mining method and system based on electronic medical record analysis

By segmenting and screening options sets in electronic medical record data, using the Apriori algorithm to define the left and right position method, the problem of traditional association rule algorithms generating useless rules in medical data is solved, improving the efficiency of rule mining and reducing data redundancy. Effective exploration of post-operative infection rules has been achieved.

CN114385718BActive Publication Date: 2024-11-15XIANGYA HOSPITAL CENT SOUTH UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210071917.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-21
Publication Date
2024-11-15
Estimated Expiration
2042-01-21

AI Technical Summary

Technical Problem

The traditional Apriori association rule algorithm will generate a large number of meaningless rules when processing medical data with large dimensions, resulting in serious data redundancy and unable to effectively improve the efficiency of association rules.

Method used

By dividing the features in the electronic medical record data into the left item set and the right item set, traversing the data set to determine the support and confidence, Apriori's association rule mining algorithm that limits the left and right item positions is used to eliminate the item set that does not meet the conditions, and avoid The emergence of useless rules improves the efficiency of association rules.

Benefits of technology

While reducing the running time of the algorithm and memory usage, the efficiency of rule mining is improved, and the rules that may cause infection after surgery can be effectively mined, overcome the limitations of traditional methods, and produce meaningful rules.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114385718B_ABST
    Figure CN114385718B_ABST
Patent Text Reader

Abstract

The present invention relates to an implicit rule mining method and system based on electronic medical record analysis. The method comprises: dividing the features extracted from the electronic medical record data set into a left item set and a right item set; traversing the electronic medical record data set according to all non-empty subsets in the left item set, determining the support of each non-empty left item subset; and saving the non-empty left item subsets whose support is greater than or equal to the support threshold, determining the first set; saving all non-empty subsets of the right item set, determining the second set; performing a direct product on the first set and the second set, obtaining all combinations of the left item subset and the right item subset, and determining the third set; traversing the electronic medical record data set, determining the support of each combination of the third set; and determining the combination whose support is greater than or equal to the support threshold as a frequent item set; determining the confidence of each frequent item set; and determining a strong association rule according to the frequent item set whose confidence is greater than or equal to the confidence threshold. The present invention can improve the efficiency of association rules.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of statistical analysis, and in particular to an implicit rule mining method and system based on electronic medical record analysis. Background Art

[0002] With the widespread popularization of electronic medical records and the rapid development of information technology, the medical field has entered a new information age, resulting in an increasing amount of medical data. Facing a massive amount of medical data, using data mining technology to analyze and mine it can provide new ideas and methods for medical personnel's clinical diagnosis and prognosis analysis.

[0003] Among the many data mining methods, association rule mining is one of the most widely used methods in the medical field. Association rules can be used to mine the potential rules contained in medical data, thereby obtaining valuable information. However, if the traditional Apriori association rule algorithm is directly applied to medical data with large dimensions, a large number of meaningless rules will be generated, resulting in serious data redundancy.

[0004] Therefore, an analysis method is urgently needed to improve the efficiency of association rules. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for mining implicit rules based on electronic medical record analysis, which can improve the efficiency of association rules.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] An implicit rule mining method based on electronic medical record analysis, comprising:

[0008] The features extracted from the electronic medical record data set are divided into a left item set and a right item set; the left item set is the features before the operation; the right item set is the features after the operation;

[0009] According to all non-empty subsets in the left item set, the electronic medical record data set is traversed to determine the support of each non-empty left item subset; and the non-empty left item subsets whose support is greater than or equal to the support threshold are saved to determine the first set;

[0010] All non-empty subsets of the right item set are saved to determine a second set;

[0011] Taking a direct product of the first set and the second set, obtaining all combinations of left-term subsets and right-term subsets, and determining a third set;

[0012] Traversing the electronic medical record data set, determining the support of each combination of the third set; and determining the combination with a support greater than or equal to a support threshold as a frequent item set;

[0013] Determine the confidence of each frequent item set; and determine strong association rules based on frequent item sets whose confidence is greater than or equal to the confidence threshold.

[0014] Optionally, the features extracted from the electronic medical record data set are divided into a left item set and a right item set, and the method also includes:

[0015] Access to electronic medical record datasets;

[0016] The electronic medical record data set is preprocessed; the preprocessing includes: data extraction, normalization, feature value selection and missing value processing.

[0017] Optionally, the step of determining the confidence of each frequent item set and determining a strong association rule according to the frequent item sets whose confidence is greater than or equal to the confidence threshold may further include:

[0018] Determine the lift based on strong association rules;

[0019] Determine whether the improvement is less than 1;

[0020] If the lift is less than 1, the strong association rule is invalid;

[0021] If the lift is greater than or equal to 1, the strong association rule is valid.

[0022] An implicit rule mining system based on electronic medical record analysis, comprising:

[0023] A feature division module is used to divide the features extracted from the electronic medical record data set into a left item set and a right item set; the left item set is the features before the operation; the right item set is the features after the operation;

[0024] The first set determination module is used to traverse the electronic medical record data set according to all non-empty subsets in the left item set, determine the support of each non-empty left item subset; and save the non-empty left item subsets whose support is greater than or equal to the support threshold to determine the first set;

[0025] A second set determination module, used for saving all non-empty subsets of the right item set to determine a second set;

[0026] A third set determination module, configured to perform a direct product of the first set and the second set to obtain all combinations of left item subsets and right item subsets, and determine a third set;

[0027] A frequent item set determination module is used to traverse the electronic medical record data set to determine the support of each combination of the third set; and determine the combination with a support greater than or equal to a support threshold as a frequent item set;

[0028] The strong association rule determination module is used to determine the confidence of each frequent item set; and determine the strong association rule according to the frequent item sets whose confidence is greater than or equal to the confidence threshold.

[0029] Optionally, it also includes:

[0030] An electronic medical record data set acquisition module, used to acquire an electronic medical record data set;

[0031] The preprocessing module is used to preprocess the electronic medical record data set; the preprocessing includes: data extraction, normalization, feature value selection and missing value processing.

[0032] Optionally, it also includes:

[0033] A lift determination module, used to determine the lift according to the strong association rule;

[0034] A first judgment module is used to judge whether the lifting degree is less than 1;

[0035] A first determination module for judging the result is used to determine that if the lifting degree is less than 1, the strong association rule is invalid;

[0036] The second determination module of the judgment result is used to determine that if the lifting degree is greater than or equal to 1, the strong association rule is valid.

[0037] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0038] The present invention provides an implicit rule mining method and system based on electronic medical record analysis, i.e., utilizing an association rule mining algorithm based on Apriori to limit the positions of left and right items, to achieve mining and analysis of rules that may cause infection after surgery, thereby overcoming the limitation that traditional statistical analysis methods are too dependent on prior knowledge and cannot mine implicit infection rules. The present invention achieves the purpose of limiting the positions of left and right items by splitting and screening item sets in the process of generating rules, avoids the generation of useless rules, greatly improves the efficiency of association rules, and thus improves the efficiency of the algorithm. By limiting the positions of left and right items of association rules, unqualified items and item sets are promptly eliminated during the execution of the algorithm, greatly reducing data redundancy, thereby reducing the running time and memory usage of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0040] Figure 1 A flowchart of an implicit rule mining method based on electronic medical record analysis provided by the present invention;

[0041] Figure 2 The figure is a flow chart of postoperative infection rule mining based on electronic medical records in an embodiment provided by the present invention. DETAILED DESCRIPTION

[0042] 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 only 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.

[0043] The purpose of the present invention is to provide a method and system for mining implicit rules based on electronic medical record analysis, which can improve the efficiency of association rules.

[0044] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] Figure 1 A flowchart of an implicit rule mining method based on electronic medical record analysis provided by the present invention is shown in FIG. Figure 1 As shown, the present invention provides an implicit rule mining method based on electronic medical record analysis, comprising:

[0046] The features extracted from the electronic medical record data set are divided into a left item set and a right item set; the left item set is the features before the operation; the right item set is the features after the operation;

[0047] Preoperative characteristics include but are not limited to gender, age, smoking history, whether there is hemiplegia before surgery, whether one can walk, whether there are certain symptoms (headache, dizziness, vomiting, fever, unsteady walking, urinary incontinence, impaired consciousness, slow reaction), preoperative routine blood values ​​(blood pressure, blood type, red blood cells, hemoglobin), preoperative routine cerebrospinal fluid values ​​(white blood cells, glucose, trace protein), routine coagulation values ​​(prothrombin time, international normalized ratio, activated partial thromboplastin time, thrombin time, fibrinogen, D-dimer), preoperative lumbar puncture pressure value, and surgical method.

[0048] Postoperative characteristics include, but are not limited to, whether there is intracranial infection, infection in other parts of the body, and whether complications occur.

[0049] According to all non-empty subsets in the left item set, the electronic medical record data set is traversed to determine the support of each non-empty left item subset; and the non-empty left item subsets whose support is greater than or equal to the support threshold are saved to determine the first set;

[0050] All non-empty subsets of the right item set are saved to determine a second set;

[0051] Taking a direct product of the first set and the second set, obtaining all combinations of left-term subsets and right-term subsets, and determining a third set;

[0052] Traversing the electronic medical record data set, determining the support of each combination of the third set; and determining the combination with a support greater than or equal to a support threshold as a frequent item set;

[0053] Determine the confidence of each frequent item set; and determine strong association rules based on frequent item sets whose confidence is greater than or equal to the confidence threshold.

[0054] (1) For electronic medical record transaction datasets D , from several medical records T Each medical record T can be considered as a k Features ( x 1, x 2, ..., x k ) X Among them, for X Any feature in x i There are several different eigenvalues.

[0055] (2) The set of feature items limited to the left item of the association rule is recorded as the left item set L , L ={ x 11 , x 12 , …, x 1j , x 21 , x 22 , …, x 2j , …, x m1 , x m2 , …, x mj}, each element in the set represents a specific eigenvalue. x mj Indicates m The feature ofj eigenvalues, , .

[0056] (3) The set of feature items limited to the right item of the association rule is recorded as the right item set R , R ={ x 11 , x 12 , …, x 1j , x 21 , x 22 , …, x 2j , …, x n1 , x n2 , …, x nj}, each element in the set represents a specific eigenvalue. x nj Indicates n The first j eigenvalues, , .

[0057] (4) , , that is, there are no identical features in the left item set and the right item set.

[0058] (5) L_Sub represents the set of all non-empty subsets of the left item set. L_Sub = R_Sub represents the set of all non-empty subsets of the right item set. R_Sub ={ }.

[0059] (6) S stands for storage L_Sub and R_Sub The set of itemsets resulting from the direct product of , S ={ }.

[0060] (7) Shape represents an association rule, which can be interpreted as X Produced under the conditions Y ,in , , X , Y They are respectively called the left term and right term of the association rule.

[0061] (8) Support, confidence and lift are commonly used indicators to measure the value of association rules. Support represents the probability of the data items in the association rule appearing in the entire data set, and confidence reflects the probability of the right item of the rule appearing under the condition of the left item of the rule. The higher the confidence, the stronger the correlation between the left and right items of the rule. A strong association rule must meet both the minimum support threshold minSup and the minimum confidence threshold minConf. Lift is used to measure the effectiveness of a strong association rule. If the lift is greater than 1, it is a valid strong association rule. The formulas for support, confidence and lift are as follows:

[0062] , ;

[0063] ;

[0064]

[0065] in, , represents the total transaction count in the transaction dataset, Indicates that the transaction data set contains X The transaction count of .

[0066] like Figure 2 As shown, the electronic medical record is preprocessed, mainly extracting the main characteristic information and other characteristic information of the patient from the electronic medical record and representing it as structured data.

[0067] The electronic medical record data set is preprocessed; the preprocessing includes: data extraction, normalization, feature value selection and missing value processing.

[0068] Specific steps for electronic medical record data preprocessing:

[0069] Step 1: Extract data. According to the opinions of professional doctors, determine the characteristic information that needs to be extracted from the electronic medical record, including basic information such as gender, age, personal history, preoperative symptoms, postoperative symptoms, blood routine, coagulation routine, surgical method and other clinical information.

[0070] Step 2: Unify the definition (normalization). Unify the definition of disease information that appears in the medical records with different expressions but the same meaning. For example, "cognitive impairment", "confusion of consciousness", "unresponsiveness to calls", etc. are uniformly defined as "disorder of consciousness".

[0071] Step 3: Feature value selection. For the symptom feature extracted from the medical record, if the patient meets the corresponding symptom, the feature value of the symptom is "1", otherwise the value is "2". For features with more than two forms, each form is valued in ascending order starting from "1". For example, taking hydrocephalus surgery as an example, the possible causes of hydrocephalus include congenital, tumor, cerebral hemorrhage, trauma, etc., which correspond to the feature values ​​of "1", "2", "3", "4", etc. at the time of feature value selection. At the same time, each feature is assigned a feature number, and the feature number increases in sequence from "01", "02"..., and finally the feature value of each feature is composed of the feature number and the feature value. For example, "031" represents the feature with the feature value of "1" under the feature number "03".

[0072] Step 4: Missing value processing. For feature information that cannot be obtained from the electronic medical record, it is impossible to obtain feature values ​​according to the method in the previous step. In order to ensure the authenticity of the data and the availability of the algorithm model, we assign the feature value "0" to the missing features and then combine it with the feature number to fill the missing values.

[0073] Among the extracted features, some features (age, blood pressure, blood routine, coagulation routine) are continuous numerical forms. Since continuous features have poor robustness and will interfere with the model, we need to discretize the continuous features through data classification. Data classification is mainly divided according to the clinical normal range, which is divided into three situations: within the normal range, below the normal range, and above the normal range, and takes discrete values ​​​​"1", "2", and "3" respectively.

[0074] Taking age, hemoglobin, and thrombin time as examples, specific examples of feature classification and discretization are shown in Tables 1, 2, and 3. Among them, hemoglobin is specifically illustrated by taking male as an example.

[0075] Table 1

[0076]

[0077] Table 2

[0078]

[0079] Table 3

[0080]

[0081] In order to verify the validity of the strong association rule, the confidence of each frequent item set is determined; and the strong association rule is determined according to the frequent item sets whose confidence is greater than or equal to the confidence threshold, and then the following is further included:

[0082] Determine the lift based on strong association rules;

[0083] Determine whether the improvement is less than 1;

[0084] If the lift is less than 1, the strong association rule is invalid;

[0085] If the lift is greater than or equal to 1, the strong association rule is valid.

[0086] The association rule mining algorithm based on Apriori with limited left and right item positions can be used to mine association rules with medical value from electronic medical record data. The left items of the association rules are all related symptoms, indicators, attributes and other characteristics before surgery, and the right items of the association rules are all symptoms related to postoperative infection after surgery. We can conduct in-depth analysis of the corresponding association rules based on the support, confidence and lift of each association rule.

[0087] Support is a macro-statistical analysis of the rules, that is, the proportion of the association rules in the entire data set; lift reflects the effectiveness of the association rules, so we first sort the association rules from high to low according to lift; then, according to the confidence, we analyze the influence of the relevant factors contained in the left item of the rule on the infection symptoms represented by the right item of the rule. The higher the confidence, the greater the possibility that the factor will cause postoperative infection, so as to analyze the important factors causing postoperative infection.

[0088] The present invention provides an implicit rule mining system based on electronic medical record analysis, comprising:

[0089] A feature division module is used to divide the features extracted from the electronic medical record data set into a left item set and a right item set; the left item set is the features before the operation; the right item set is the features after the operation;

[0090] The first set determination module is used to traverse the electronic medical record data set according to all non-empty subsets in the left item set, determine the support of each non-empty left item subset; and save the non-empty left item subsets whose support is greater than or equal to the support threshold to determine the first set;

[0091] A second set determination module, used for saving all non-empty subsets of the right item set to determine a second set;

[0092] A third set determination module, configured to perform a direct product of the first set and the second set to obtain all combinations of left item subsets and right item subsets, and determine a third set;

[0093] A frequent item set determination module is used to traverse the electronic medical record data set to determine the support of each combination of the third set; and determine the combination with a support greater than or equal to a support threshold as a frequent item set;

[0094] The strong association rule determination module is used to determine the confidence of each frequent item set; and determine the strong association rule according to the frequent item sets whose confidence is greater than or equal to the confidence threshold.

[0095] The present invention provides an implicit rule mining system based on electronic medical record analysis, further comprising:

[0096] An electronic medical record data set acquisition module, used to acquire an electronic medical record data set;

[0097] The preprocessing module is used to preprocess the electronic medical record data set; the preprocessing includes: data extraction, normalization, feature value selection and missing value processing.

[0098] The present invention provides an implicit rule mining system based on electronic medical record analysis, further comprising:

[0099] A lift determination module, used to determine the lift according to the strong association rule;

[0100] A first judgment module is used to judge whether the lifting degree is less than 1;

[0101] A first determination module for judging the result is used to determine that if the lifting degree is less than 1, the strong association rule is invalid;

[0102] The second determination module of the judgment result is used to determine that if the lifting degree is greater than or equal to 1, the strong association rule is valid.

[0103] The present invention utilizes the association rule method to mine the rules that may cause postoperative infection of hydrocephalus from the electronic medical record data. This method is a non-purposeful unsupervised learning method that focuses on finding the correlation between features. On the one hand, it avoids the loss of time and manpower caused by manual intervention, and on the other hand, it breaks the limitations of thinking and can mine implicit potential rules that are different from the usual prior knowledge. In view of the shortcomings and limitations of the traditional Apriori algorithm, the traditional Apriori algorithm is improved. In the process of generating rules, the item set is split and screened to achieve the purpose of limiting the position of the left and right items, avoiding the generation of useless rules, solving the problem of serious data redundancy, improving the efficiency of generating meaningful rules, and greatly reducing the execution time and memory usage of the algorithm.

[0104] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0105] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. An implicit rule mining method based on electronic medical record analysis, It is characterized in that include: The features extracted from the electronic medical record data set are divided into a left item set and a right item set; the left item set is the features before the operation; The right item set is the post-operative features; According to all non-empty subsets in the left item set, traverse the electronic medical record data set to determine the support of each non-empty left item subset; And save the non-empty left item subsets whose support is greater than or equal to the support threshold to determine the first set; All non-empty subsets of the right item set are saved to determine a second set; Taking a direct product of the first set and the second set, obtaining all combinations of left-term subsets and right-term subsets, and determining a third set; Traversing the electronic medical record data set to determine the support of each combination of the third set; And the combination whose support is greater than or equal to the support threshold is determined as a frequent item set; Determine the confidence level of each frequent itemset; And determine the strong association rules based on the frequent item sets whose confidence is greater than or equal to the confidence threshold; Determining the confidence of each frequent item set; And determine the strong association rules based on the frequent item sets whose confidence is greater than or equal to the confidence threshold, and then include: Determine the lift based on strong association rules; Determine whether the improvement is less than 1; If the lift is less than 1, the strong association rule is invalid; If the lift is greater than or equal to 1, the strong association rule is valid.

2. According to claim 1, a method for mining implicit rules based on electronic medical record analysis, It is characterized in that The features extracted from the electronic medical record dataset are divided into left item sets and right item sets, which also include: Access to electronic medical record datasets; The electronic medical record data set is preprocessed; the preprocessing includes: data extraction, normalization, feature value selection and missing value processing.

3. An implicit rule mining system based on electronic medical record analysis, It is characterized in that include: A feature division module is used to divide the features extracted from the electronic medical record data set into a left item set and a right item set; the left item set is the features before the operation; The right item set is the post-operative features; The first set determination module is used to traverse the electronic medical record data set according to all non-empty subsets in the left item set, and determine the support of each non-empty left item subset; And save the non-empty left item subsets whose support is greater than or equal to the support threshold to determine the first set; A second set determination module, used for saving all non-empty subsets of the right item set to determine a second set; A third set determination module, configured to perform a direct product of the first set and the second set to obtain all combinations of left item subsets and right item subsets, and determine a third set; A frequent item set determination module, used to traverse the electronic medical record data set and determine the support of each combination of the third set; And the combination whose support is greater than or equal to the support threshold is determined as a frequent item set; Strong association rule determination module, used to determine the confidence of each frequent item set; And determine the strong association rules based on the frequent item sets whose confidence is greater than or equal to the confidence threshold; A lift determination module, used to determine the lift according to the strong association rule; A first judgment module is used to judge whether the lifting degree is less than 1; A first determination module for judging the result is used to determine that if the lifting degree is less than 1, the strong association rule is invalid; The second determination module of the judgment result is used to determine that if the lifting degree is greater than or equal to 1, the strong association rule is valid.

4. The implicit rule mining system based on electronic medical record analysis according to claim 3, It is characterized in that Also includes: An electronic medical record data set acquisition module, used to acquire an electronic medical record data set; A preprocessing module, used for preprocessing the electronic medical record data set; The preprocessing includes: data extraction, normalization, feature selection and missing value processing.