Construction method and system of old people multi-disease co-management prediction model based on big data and co-management system

Abstract

The invention provides a big data-based construction method of an old people multi-disease co-management prediction model and a co-management system. The method comprises the steps of collecting group data related to old people syndrome pathogenesis factors; extracting relation semantics and attribute semantics from the collected group data; constructing a decision table data set based on the obtained relation semantics and attribute semantics; dynamically performing attribute reduction on the decision table data set based on the decision entropy to obtain a feature reduction data set; and taking the reduced decision table data set as a training data set to train a machine learning model, and constructing and training a disease prediction model. Attribute reduction is dynamically carried out on the decision table data set based on the decision entropy, and the accuracy of classification of different data information and the efficiency of model training can be greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data processing, and in particular to a method for constructing a multi-disease co-management system for the elderly based on big data and a prediction model. Background Art

[0002] The elderly are the group with the highest prevalence and incidence of chronic diseases. Due to the aging of tissues and organs, the elderly are prone to degenerative and metabolic diseases. Due to the weakened immune function, they are prone to infection and tumors. In addition, when one system is diseased, it can cause another system to be diseased through the mutual influence between systems. The manifestation of coexistence of multiple diseases is mainly the coexistence of multiple system diseases. For example, an elderly person may suffer from hypertension, coronary heart disease, diabetes, urinary tract infection and other diseases at the same time.

[0003] Co-management of multiple diseases in the elderly is a comprehensive health management strategy. It is based on the analysis of health and medical big data of the elderly with multiple diseases, and can achieve early risk identification of specific diseases in the elderly and effectively manage multiple common chronic diseases in the elderly. It can improve the efficiency and quality of health management of the elderly with multiple diseases and effectively reduce the socioeconomic burden of the elderly with multiple diseases.

[0004] A common method of achieving multi-disease co-management for the elderly is to use big data technology to analyze the collected data, identify disease patterns, risk factors and potential health problems, and conduct disease risk assessments on the elderly based on the analysis results to predict possible health problems in the future. Data collection includes medical test report data. For different specific diseases, such as hypertension, coronary heart disease, diabetes, urinary tract infection, etc., the format and content of medical test reports may be significantly different, which will bring great difficulties and challenges to the data classification for multi-disease co-management. In addition, most of the current data classification methods are based on static methods. Each time a new data object is added, the data model needs to be retrained from scratch, which will seriously reduce the efficiency of model training. Summary of the invention

[0005] The present invention provides a method for constructing a prediction model for co-management of multiple diseases in the elderly based on big data and a co-management system for multiple diseases in the elderly based on big data.

[0006] In a first aspect, the present invention provides a method for constructing a prediction model for multi-disease co-management of the elderly based on big data, characterized in that the method comprises:

[0007] To collect cohort data related to the etiology of geriatric syndromes;

[0008] Extract relational and attribute semantics from the collected group data;

[0009] Construct a decision table data set based on the acquired relational semantics and attribute semantics; dynamically simplify the attributes of the decision table data set based on decision entropy to obtain a feature-reduced data set;

[0010] The simplified decision table data set is used as the training data set to train the machine learning model, and the disease prediction model is constructed and trained.

[0011] In one embodiment of the present invention, dynamically simplifying the attributes of decision table data based on decision entropy includes: constructing a decision rule tree; and dynamically updating the rule tree by comparing the newly added object with the existing rules based on the decision entropy.

[0012] In one embodiment of the present invention, the comparison of the newly added object with the existing rules includes at least one of the following situations: the original simplified attribute set can distinguish the new object, and the conditional attribute can also distinguish the new object; the original simplified attribute set cannot distinguish the new object, and the conditional attribute set can distinguish the new object; the original simplified attribute set cannot distinguish the new object, and the conditional attribute set cannot distinguish the new object.

[0013] In one embodiment of the present invention, after a new object is added, if the original simplified attribute set can distinguish the new object and the conditional attribute can also distinguish the new object, the simplified attribute remains unchanged; if the original simplified attribute set cannot distinguish the new object and the conditional attribute set can distinguish the new object, the simplification is recalculated according to the decision entropy; if the original simplified attribute set cannot distinguish the new object and the conditional attribute set cannot distinguish the new object, the simplification remains unchanged.

[0014] In one embodiment of the present invention, it also includes using evaluation indicators to evaluate the performance of the model and adjusting and optimizing the model according to the evaluation results.

[0015] In one embodiment of the present invention, the machine learning model is a recurrent neural network model or a convolutional neural network model.

[0016] In one embodiment of the present invention, the received input values ​​of geriatric syndrome risk factors include parameter values ​​of waist circumference, blood pressure, blood sugar, high-density lipoprotein cholesterol, insulin and neutral fat.

[0017] On the other hand, the present invention provides a multi-disease co-management system for the elderly based on big data, comprising:

[0018] An input unit receives input values ​​of geriatric syndrome risk factors and provides them to an analysis and processing unit;

[0019] The analysis and processing unit uses the disease prediction model to predict the risk and type of disease;

[0020] The execution unit gives a corresponding health management strategy according to the prediction results of the analysis and processing unit;

[0021] Wherein, the method for constructing the disease prediction model is as follows:

[0022] To collect cohort data related to the etiology of geriatric syndromes;

[0023] Extract relational and attribute semantics from the collected group data;

[0024] Construct a decision table data set based on the acquired relational semantics and attribute semantics; dynamically simplify the attributes of the decision table data set based on decision entropy to obtain a feature-reduced data set;

[0025] The simplified decision table data set is used as the training data set to train the machine learning model, and the disease prediction model is constructed and trained.

[0026] In one embodiment of the present invention, the multiple diseases include at least one of hypertension, coronary heart disease, diabetes, and urinary tract infection.

[0027] In one embodiment of the present invention, the corresponding health management strategy is given according to the prediction results of the analysis and processing unit, including blood pressure control, blood sugar control, weight management, smoking cessation and alcohol restriction or mental health intervention according to the prediction results.

[0028] It has the following beneficial effects:

[0029] The prediction model construction method provided by the present invention dynamically simplifies the attributes of the decision table data set, which can greatly improve the accuracy of data classification and the efficiency of model training.

[0030] The prediction model construction method provided by the present invention recursively constructs a decision rule tree from top to bottom, the calculation is intuitive, and the efficiency of data set processing can be greatly improved.

[0031] The effects of the present invention are not limited to the above-mentioned effects, and should be understood to include all effects that can be inferred from the detailed description of the present invention or the scope of patent application.

[0032] Figure 1 Diagram of the data processing flow of the method for constructing a prediction model for co-management of multiple diseases in the elderly based on big data.

[0033] Figure 2 The block diagram of the multi-disease co-management system for the elderly based on big data is shown.

[0034] The advantages and features of the present invention, and methods of achieving them, will become clear through the embodiments described in detail with reference to the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, and can be implemented in different various forms, but these embodiments provide a complete disclosure of the present invention, and in order to fully inform the ordinary technicians in this technical field of the scope of the invention, the present invention is only defined by the scope of the claims. The same figure symbols refer to the same components throughout the specification.

[0035] When describing the embodiments of the present invention, if it is considered that the detailed description of known functions or structures unnecessarily obscures the main points of the present invention, such detailed description will be omitted. Moreover, the terms mentioned below are defined in consideration of the functions of the embodiments of the present invention and may vary according to the intention or convention of the user or operator. Therefore, these definitions should be determined based on the contents of the entire specification.

[0036] Figure 1 It is a diagram showing the data processing flow of the method for constructing a multi-disease co-management prediction model for the elderly based on big data.

[0037] In step S110, group data related to the pathogenic factors of geriatric syndrome are collected. In step S110, the group data is the detection index data that can reflect a specific disease condition and the corresponding basic information of the personnel, medical history, medication records and other information.

[0038] In step S120, relational semantics and attribute semantics are extracted from the collected group data. Relational semantics is the probabilistic relationship between different indicator parameters and disease diagnosis results. For example, low insulin parameters will lead to a high probability of diabetes. Attribute semantics represents different types of data information.

[0039] In step S130, a decision table data set is constructed based on the acquired relational semantics and attribute semantics; and attributes of the decision table data set are dynamically simplified based on decision entropy to obtain a feature-reduced data set.

[0040] Specifically, the decision table is a special and very important type of knowledge expression system. Specifically, the decision table is also a special information table, which indicates how the decision should be made when certain conditions are met. Therefore, the definition of the decision table is:

[0041] A decision table information system (abbreviated as decision table) S = <U, R, V, f>, where U is the set of all objects, also known as the domain, R = C∪D represents the attribute set, where subsets C and D represent the condition attribute set and decision attribute set respectively. V=V r Represents a set of attribute values, V ris the value domain of attribute r, that is, the range of attribute values ​​of attribute r∈R, and f:U×R→V is an information function that specifies the attribute value of each object x in the domain U.

[0042] Rough set theory is a mathematical tool to describe incompleteness and uncertainty. Rough set theory is based on indistinguishable relations and provides a model of knowledge expression system. It also uses accurate upper and lower approximate sets to approximate imprecise objects from the perspective of classification, giving imprecise knowledge a clear mathematical meaning. Rough set theory can not only effectively analyze and process various types of incomplete information such as imprecise, inconsistent, and incomplete, but also discover the implicit knowledge and potential laws from it.

[0043] When using rough sets to process decision tables, each attribute value in the decision table is required to be discrete data. Therefore, it is necessary to discretize it first, that is, to discretize those conditional attributes whose value domains are continuous or to abstract the already discrete data to obtain higher discrete values. Then, attribute simplification is performed on the discretized information system. Since not every attribute in the information system is equally important, and some attributes in some decision rules have no effect on the decision rules of the entire information system, it is necessary to process each attribute in the information system according to a certain strategy to complete the decision table simplification process and obtain the minimum attribute set. That is, the simplified decision table has the same function as the decision table before simplification, but the conditional attribute set is minimized. This is attribute simplification.

[0044] In rough set theory, decision entropy is based on the invariant standard of decision-making ability, so it can accurately and objectively reflect the essence of decision-making ability.

[0045] Dynamically reducing the attributes of the decision table data set based on decision entropy includes:

[0046] Construct a decision rule tree; based on this rule tree, dynamically update the rule tree by comparing the newly added objects with the existing rules in combination with decision entropy. Specifically, attribute simplification is performed on the decision table data set based on decision entropy; after a new object is added, if the original simplified attribute set B can distinguish the new object and the conditional attribute C can also distinguish the new object, the simplified attribute remains unchanged; if the original simplified attribute set B cannot distinguish the new object and the conditional attribute set C can distinguish the new object, the simplification is recalculated based on decision entropy; if the original simplified attribute set B cannot distinguish the new object and the conditional attribute set C cannot distinguish the new object, the simplification remains unchanged.

[0047] The decision rule tree refers to a tree structure that represents a decision set. These decision sets generate decision rules by classifying data sets. If S = (U, C, D, V, f) is a consistent decision table, then each leaf node of the decision tree only corresponds to objects of the same decision class, that is, each leaf node corresponds to a deterministic decision rule, and its credibility (conditional probability distribution of the object) is equal to 1. Otherwise, some leaf nodes of the decision tree correspond to objects of different decision classes. Such leaf nodes correspond to uncertain decision rules, and their feasibility is less than 1. Therefore, the attribute reduction method is designed based on the attribute importance of decision entropy as inspiration information. First, starting from the empty tree T, conditional attributes are gradually added, and the minimum knowledge rough entropy is selected. The attributes covering the most decision classification objects are used as branch nodes, and the decision tree is recursively constructed from top to bottom; then, according to the idea of ​​block processing, the useful decision rules implicit in the decision table are extracted with as few attributes as possible; finally, all redundant conditional attributes that do not affect the expression of the rules are deleted to simplify the decision rules. The specific operation steps of this method are described as follows:

[0048] Decision rule tree construction

[0049] Input: Decision table S = (U, C, D, V, f);

[0050] Output: decision rule tree T and decision rule set.

[0051] Step 1: Merge the same objects in the decision table S;

[0052] Step 2: Initialization T is an empty tree;

[0053] Step 3: For any attribute a∈CB, calculate E(D|B∪{a});

[0054] Step 4: Select the attribute a that minimizes E(D|B∪{a}) as the root node (or branch node) of the decision tree T

[0055] If there are multiple attributes that simultaneously minimize E(D|B∪{a}), then select an attribute a from them so that its dependency on B is γ B∪{a} (D) Maximum, if there are still multiple attributes that make γ B∪{a} (D) When the maximum value is reached, the attribute with the highest order is selected; Step 5: Classify U using the selected attribute a, that is, calculate U / {a}={U1,U2,...,U t}, start to build the sub-decision table (i.e. the branch of the decision tree) S i =(U i ,C,D,V,f), where i=1,2,...,t;

[0056] Step 6: If branch S iAll objects U in (i=1,2,...,t) i With the same decision attribute value, in branch S i Generate a leaf node, identify its decision attribute value, traverse a path from the root to the leaf node, and generate the corresponding decision rule

[0057] If any non-leaf node in the decision rule is removed, in S i Then continue to remove the second one.

[0058] 3, ... non-leaf nodes until they cannot be found in S i The only expression

[0059] else B=B∪{a};

[0060] Step 7: if B = C or U is completely classified by the decision tree branch, then output the decision tree T and the decision rule set.

[0061] elsegoto Step3

[0062] end;

[0063] In the decision table, the credibility and object coverage of decision rules are important indicators to measure decision-making ability. Based on the rough entropy of knowledge, a new rough entropy definition, decision entropy, is given, and its attribute importance is defined; then the decision entropy of the conditional attribute subset is used to measure its importance to the decision classification, and the attribute with the smallest decision entropy and covering the most decision classification objects is selected as the branch node, and the decision tree is recursively constructed from top to bottom; finally, the decision tree is traversed to simplify the obtained decision rules.

[0064] Incremental knowledge acquisition is a dynamic knowledge acquisition method that directly processes newly added objects by comparing them with existing knowledge without recalculating all data sets. Therefore, the core of the incremental algorithm is to compare and distinguish in multiple situations. There are three situations when adding new objects:

[0065] (1) The rules in the existing rule set can identify the new object and have the same decision value (match);

[0066] (2) The rules in the existing rule set can identify the new object, but the decision values ​​are different (contradiction);

[0067] (3) The rules in the existing rule set cannot recognize the new object.

[0068] First, in order to quickly match the newly added objects, a rule tree is constructed for the rule set, and each path from the root node to the leaf node represents a rule; secondly, based on this rule tree, combined with decision entropy, the rule tree is dynamically updated by comparing the newly added objects with the existing rules to achieve dynamic updating of knowledge.

[0069] Incremental rule acquisition method input: known decision table S = <U, A = C ∪ D, V, f>, determine the rule set R (R corresponding to the decision table S' = <U', A' = C' ∪ D', V', f'>), the newly added object u

[0070] Output: New uncertainty rule set NDR Step 1: (Initialization) Original rule set ODR = R; NDR = R; C' = {c1, c2, ... c m};

[0071] number = 0 ( / / number represents the total number of objects in R that are in the same coverage as u); Step 2: (Construct rule tree)

[0072] For the rule set R, call the above decision rule tree method to construct a decision rule tree and return the rule tree Rules_Tree;

[0073] Step 3: Read the data in u in sequence according to the attributes contained in C'∪D'.

[0074] Determine the relationship between u and the rules in R, and match the rule tree layer by layer from top to bottom.

[0075] Case 1 (new object matches the rule tree exactly):

[0076] Matching from the root node to a leaf node means that both the condition attribute and the decision attribute can be fully matched, and there is no need to update the rule tree, and go to Step 8;

[0077] Case 2 (new object conflicts with the rule tree):

[0078] If the node that matches the previous node of a leaf node is not equal to the value on the leaf node, that is, the corresponding condition attribute is completely matched, but the decision attribute is not matched (contradictory), then:

[0079] (1) Obtain the contradictory rule r';

[0080] (2) Find the object set α corresponding to the rule in S, and add attribute values ​​to the attributes reduced by r' to distinguish the newly added object u from the object set α;

[0081] (3) Delete rule r' from the rule tree, insert r' after adding the attribute value into the rule tree, and go to Step 7;

[0082] Case 3 (a new object generates a new rule):

[0083] u cannot find a matching rule, that is, u generates a new rule r, R = R ∪ {r}; merge object u into the original decision table to obtain a new decision table s.

[0084] Step 4: Simplify the values ​​of the new decision table s (the method of value simplification is to repeat Step 5 to Step 6);

[0085] Step 5: For any a i ∈C, calculate -E({a i}→D)

[0086] Step 5.1: Select -E({a i}→D) The largest attribute a i As the most important attribute; if multiple attributes of E({a i}→D) If the values ​​are the same, the attribute with the higher order is selected;

[0087] Step 5.2: Let Aset = Aset∪{a i}, if Aset≠C, go to Step 5;

[0088] Step 5.3: Output Aset = {a1, a2, ... a m};

[0089] Step 6: In the attribute set Aset-{a k}(k=1...i) according to the tolerance relation to decompose (U'∪{u}) to obtain the cover about u.

[0090] Step 6.1: If the number of elements in u's coverage is 1, delete u in attribute a. k The attribute value on is recorded as NULL;

[0091] Step 6.2: If the number of elements in u's coverage is greater than 1, calculate the confidence Conf of u. If Conf≠Conf' (Conf' is the number of elements in u's coverage without a k The previous confidence level), then attribute column a k The attribute value must be retained; if Conf=Conf', then calculate the decision entropy of u. If E≤E' (E' is the value of removing a k

[0092] The decision entropy before), then the attribute column a k The attribute value can be deleted and recorded as NULL, otherwise the attribute value is retained;

[0093] Step 7: Construct rule r based on u. If NDR=R∪{r}, update the rule tree;

[0094] Step 8: Return the new rule tree NDR;

[0095] In step S140, the simplified decision table data set is used as a training data set to train the machine learning model, and a disease prediction model is constructed and trained.

[0096] Specifically, choose a suitable machine learning model, which may include logistic regression, random forest, support vector machine, convolutional neural network, recurrent neural network, etc.;

[0097] Use the reduced dataset to train the selected model. At this stage, it is usually necessary to divide the dataset into a training set and a validation set to evaluate the performance of the model;

[0098] Use appropriate evaluation metrics (such as precision, recall, F1 score, ROC-AUC, etc.) to evaluate the performance of the model;

[0099] Adjust and optimize the model based on the evaluation results, which may include adjusting model parameters, using more complex model structures, etc.; deploy the trained model to actual applications for disease prediction;

[0100] Continuously monitor the performance of the model and update and maintain the model based on new data and feedback.

[0101] Figure 2 It is a block diagram showing the multi-disease co-management system for the elderly based on big data.

[0102] like Figure 2 As shown, it includes: an input unit 201, which receives input values ​​of factors of elderly syndrome, which include index parameter values ​​such as waist circumference, blood pressure, blood sugar, high-density lipoprotein (HDL) cholesterol, insulin and neutral fat, and provides them to an analysis and processing unit 202; the analysis and processing unit 202 uses the model constructed as claimed in claims 1-4 to predict the risk and type of disease; the execution unit 203 gives corresponding health management strategies according to the results predicted by the analysis and processing unit, sets reasonable health management goals according to the specific circumstances of the elderly, such as blood pressure control, blood sugar control, weight management, etc., and can also provide guidance and intervention in diet, exercise, smoking cessation, alcohol restriction, mental health, etc., to improve the lifestyle of the elderly.

[0103] The above embodiments are merely illustrative of the principles and effects of the present application and are not intended to limit the present application. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed in the present application shall still be covered by the claims of the present application.

Claims

1. A method for constructing a prediction model for multi-disease co-management of the elderly based on big data, comprising: To collect cohort data related to the etiology of geriatric syndromes; Extract relational and attribute semantics from the collected group data; Construct a decision table data set based on the acquired relational semantics and attribute semantics; Based on decision entropy, the attributes of the decision table data set are dynamically simplified to obtain a feature-reduced data set; The simplified decision table data set is used as the training data set to train the machine learning model, and the disease prediction model is constructed and trained.

2. The model building method according to claim 1, characterized in that: Dynamically simplifying the attributes of decision table data based on decision entropy includes: constructing a decision rule tree; based on the rule tree, dynamically updating the rule tree by comparing the newly added object with the existing rules in combination with the decision entropy.

3. The model building method according to claim 1, characterized in that: The comparison of the newly added object with the existing rules includes at least one of the following situations: the original simplified attribute set can distinguish the new object, and the conditional attribute can also distinguish the new object; the original simplified attribute set cannot distinguish the new object, and the conditional attribute set can distinguish the new object; The original reduced attribute set cannot distinguish the new object, and the conditional attribute set cannot distinguish the new object either.

4. The model building method according to claim 3, after a new object is added, if the original simplified attribute set can distinguish the new object and the conditional attribute can also distinguish the new object, the simplified attribute remains unchanged; If the original reduced attribute set cannot distinguish the new object, but the conditional attribute set can distinguish the new object, the reduction is recalculated based on the decision entropy; If the original reduced attribute set cannot distinguish the new object, and the conditional attribute set cannot distinguish the new object, the reduction remains unchanged.

5. The model building method as described in claim 1 also includes using evaluation indicators to evaluate the performance of the model and adjusting and optimizing the model according to the evaluation results.

6. The model building method as described in claim 1, wherein the machine learning model is a recurrent neural network model or a convolutional neural network model.

7. The model building method as described in claim 1, wherein the input values ​​of the geriatric syndrome risk factors include parameter values ​​of abdominal circumference, blood pressure, blood sugar, high-density lipoprotein cholesterol, insulin and neutral fat.

8. A multi-disease co-management system for the elderly based on big data, comprising: An input unit receives input values ​​of geriatric syndrome risk factors and provides them to an analysis and processing unit; The analysis and processing unit uses the disease prediction model to predict the risk and type of disease; The execution unit gives a corresponding health management strategy according to the prediction results of the analysis and processing unit; Wherein, the method for constructing the disease prediction model is as follows: To collect cohort data related to the etiology of geriatric syndromes; Extract relational and attribute semantics from the collected group data; Construct a decision table data set based on the acquired relational semantics and attribute semantics; dynamically simplify the attributes of the decision table data set based on decision entropy to obtain a feature-reduced data set; The simplified decision table data set is used as the training data set to train the machine learning model, and the disease prediction model is constructed and trained.

9. The multi-disease co-management system as claimed in claim 8, wherein the multi-diseases include at least one of hypertension, coronary heart disease, diabetes, and urinary tract infection.

10. The multi-disease co-management system as described in claim 8, wherein the corresponding health management strategy is given according to the prediction results of the analysis and processing unit, including blood pressure control, blood sugar control, weight management, smoking cessation and alcohol restriction or mental health intervention according to the prediction results.

Images