CN112366003A - Adverse reaction monitoring method based on clinical real world time series elements - Google Patents

Adverse reaction monitoring method based on clinical real world time series elements Download PDF

Info

Publication number
CN112366003A
CN112366003A CN202011125974.4A CN202011125974A CN112366003A CN 112366003 A CN112366003 A CN 112366003A CN 202011125974 A CN202011125974 A CN 202011125974A CN 112366003 A CN112366003 A CN 112366003A
Authority
CN
China
Prior art keywords
adverse reaction
information
medication
time series
reaction monitoring
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202011125974.4A
Other languages
Chinese (zh)
Inventor
刘宇云
王雷
张奥萌
吕龙挺
李涛
吕翰林
方欢
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
HUBEI HUADA GENE Research Institute
Original Assignee
Wuhan Bgi Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Wuhan Bgi Technology Co ltd filed Critical Wuhan Bgi Technology Co ltd
Priority to CN202011125974.4A priority Critical patent/CN112366003A/en
Publication of CN112366003A publication Critical patent/CN112366003A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H80/00—ICT specially adapted for facilitating communication between medical practitioners or patients, e.g. for collaborative diagnosis, therapy or health monitoring
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/284—Lexical analysis, e.g. tokenisation or collocates
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
    • G16H20/10—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to drugs or medications, e.g. for ensuring correct administration to patients
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders

Abstract

The invention relates to an adverse reaction monitoring method based on clinical real world time series elements. It includes: acquiring medication information of a patient, wherein the medication information is obtained based on a medication text and a time series characteristic vector, and the medication text comprises used medicine information arranged according to a time sequence, corresponding symptom information and body index value change information; acquiring adverse reaction information of the patient, wherein the adverse reaction information indicates whether the patient has an adverse reaction or the type of the occurred adverse reaction based on the medication information; and training a preset machine learning model by using the medication information as a sample and the adverse reaction information as a label to obtain an adverse reaction monitoring model. According to the technical scheme, whether adverse reactions occur or which types of adverse reactions occur in the current medication stage or after the change of the medicine of the patient can be accurately predicted, so that corresponding preparation is made in advance, and the health of the patient is guaranteed.

Description

Adverse reaction monitoring method based on clinical real world time series elements
Technical Field
The invention relates to the technical field of computer application, in particular to an adverse reaction monitoring method based on clinical real world time series elements.
Background
Patients may require the use of multiple medications while undergoing treatment. Because the action mechanisms of different drugs are different, or the patients have different constitutions, the responses of different patients to the same drug are different, and the responses of the same patients to different drugs are different. In the long-term administration process, different medicines can be used according to treatment needs, and the administration can be adjusted according to the physical condition of a patient, at the moment, adverse reactions can occur to the patient due to the adjustment of the medicines, and if the patient waits for serious adverse reactions to be treated, the health of the patient can be seriously affected.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides an adverse reaction monitoring method based on clinical real-world time series elements.
In a first aspect, the invention provides a method for constructing an adverse reaction monitoring model based on clinical real-world time series elements, which comprises the following steps:
acquiring medication information of a patient, wherein the medication information is obtained based on a medication text and a time series characteristic vector, and the medication text comprises used medicine information arranged according to a time sequence, corresponding symptom information and body index value change information;
acquiring adverse reaction information of the patient, wherein the adverse reaction information indicates whether the patient has an adverse reaction or the type of the occurred adverse reaction based on the medication information;
and training a preset machine learning model by using the medication information as a sample and the adverse reaction information as a label to obtain a medication adverse reaction monitoring model.
Further, the acquiring medication information of the patient comprises:
acquiring the medication text;
converting the medication text into a text vector based on a word space matrix;
and acquiring a time sequence feature vector, and fusing the text vector and the time sequence feature vector.
Further, the machine learning model comprises a recurrent neural network, an attention distribution model, a feedforward network layer and a normalization function; the output of the recurrent neural network is used as the input of the attention degree distribution model, the output of the attention degree distribution model is used as the input of the feedforward network layer, and the output of the feedforward network layer is used as the input of the normalization function.
Further, the recurrent neural network is a bidirectional recurrent neural network, and a filtering mechanism is embedded in the bidirectional recurrent neural network.
Further, the feed-forward network layer comprises:
F＝WFP+bF，F^＝ψ(F)；
wherein the content of the first and second substances,p represents the output of the attention distribution model, WFWeight of a parameter representing the feedforward network layer, bFRepresents the offset of the feedforward network layer, F represents the output of the feedforward network layer that has not been subjected to the activation function psi (·), and F ^ represents the output of the feedforward network layer after being subjected to the activation function psi (·).
Further, the output of the normalization function represents two-classification output with or without adverse reaction, or multi-classification output with different types of adverse reaction.
In a second aspect, the invention provides an adverse reaction monitoring model construction device based on clinical real-world time series elements, which comprises a memory and a processor; the memory for storing a computer program; the processor is used for realizing the adverse reaction monitoring model construction method based on clinical real world time series elements when the computer program is executed.
In a third aspect, the present invention provides a method for monitoring adverse reactions based on clinical real-world time series elements, comprising the steps of:
inputting the medication information of the calibrated patient into the medication adverse reaction monitoring model constructed by the adverse reaction monitoring model construction method based on clinical real world time series elements, wherein the medication information is obtained based on a medication text and a time series characteristic vector, and the medication text comprises the medication information arranged according to time sequence and corresponding symptom information and body index value change information;
and taking the output of the adverse drug reaction monitoring model as prediction data for indicating the calibrated patient to generate adverse reaction.
In a fourth aspect, the present invention provides an adverse reaction monitoring device based on clinical real-world time series elements, the device comprising a memory and a processor; the memory for storing a computer program; the processor, when executing the computer program, is configured to implement the method for monitoring adverse reactions based on clinical real-world time series elements as described above.
In a fifth aspect, the present invention provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements a method for constructing an adverse reaction monitoring model based on clinical real-world time series elements as described above, or implements a method for monitoring an adverse reaction based on clinical real-world time series elements as described above.
The adverse reaction monitoring model construction and monitoring method, device and storage medium based on clinical real world time series elements have the advantages that time series characteristics of a series of elements such as medicines, symptoms and body index value changes in the clinical real world are brought into model learning, so that the model is stronger and more sensitive. Meanwhile, different elements are distributed and spliced according to the time sequence, so that the flexibility is higher, the incorporated different elements can be conveniently expanded, and the accuracy of model classification is further ensured. The adverse reactions of the current medication stage of the patient or the next medication stage after the medication of the patient is changed can be monitored and predicted, and whether the adverse reactions occur or which kind of adverse reactions occur can be accurately predicted based on the constructed model, so that corresponding preparation is made in advance, and the health of the patient is ensured.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, a brief description will be given below to the drawings required for the description of the embodiments or the prior art, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without inventive exercise.
FIG. 1 is a schematic flow chart of a method for constructing an adverse reaction monitoring model based on clinical real-world time series elements according to an embodiment of the present invention;
FIG. 2 is a schematic flow chart of the adverse reaction monitoring method based on clinical real-world time series elements according to the embodiment of the present invention.
Detailed Description
The principles and features of this invention are described below in conjunction with the following drawings, which are set forth by way of illustration only and are not intended to limit the scope of the invention.
As shown in fig. 1, the method for constructing an adverse reaction monitoring model based on clinical real-world time series elements according to the embodiment of the present invention includes the following steps:
s11, acquiring medication information of the patient, wherein the medication information is obtained based on a medication text and the time series characteristic vector, and the medication text comprises the use medicine information arranged in time sequence and corresponding symptom information and body index value change information.
Specifically, the medication text data source may be patient basic information (sex, age, etc.), past history, examination indexes of each item, medication record, post-medication performance record, symptom record, adverse reaction record (conclusion, grading, description), and the like recorded by the system in the hospital.
First, the data extracted from the real world are temporally concatenated into a description text with a time axis as an extension direction. The description text is mainly formed by splicing three parts according to time distribution, namely, medicines, physical sign symptoms and body index value change are used. In addition, other elements can be added according to the actual data situation for distribution splicing.
Time-series based description of the use of drugs:
for example, the medication record of Zhang III patient is extracted from the data in hospital, and the medication record is sorted by time as follows: drug A, drug B and drug C.
Time-series-based description of the signs:
for example, the sign and symptom records of Zhang III of the patient are extracted from the data in the hospital, and are sorted by time as follows: symptom a, symptom B, and symptom C.
Description of body index value change based on time series:
for example, a body index value record of Zhang III of a patient is extracted from the data in a hospital, and the body index value record is sorted by time as follows: the index A is increased, the index B is decreased, and the index C is unchanged. The body index value change description needs to be standardized to distinguish different types of changes.
It should be noted that the time-ordered scale of the elements is consistent.
In summary, the descriptions in the above sections can be extended according to the time axis, and the mixed distribution splicing is performed according to the time sequence. One possible stitching result is:
the A index of the A symptom of the drug A is increased, the B index of the B symptom of the drug B is decreased, and the C index of the C symptom of the drug C is not changed.
S12, obtaining adverse reaction information of the patient, wherein the adverse reaction information indicates whether the patient has an adverse reaction or the type of the occurred adverse reaction based on the medication information.
Specifically, large text data such as medical records and ward-round records in a hospital are subjected to structured processing, and by combining methods such as entity identification, relation extraction and rule association matching, entities such as diagnosis, positive modification, negative modification, change modification, parts, symptoms, medicines (prescriptions), diseases, disease courses, examination values, examination names, treatment methods and groups, and conclusion, classification and other information of adverse reactions are obtained from the large text. The time-series-based medication, symptom record and body index value change are associated to corresponding adverse reaction conclusion and grading.
It should be noted that steps S11 and S12 are not in strict sequence, and the step of acquiring the relevant information may be performed in parallel.
And S13, training a preset machine learning model by taking the medication information as a sample and taking the adverse reaction information as a label to obtain a medication adverse reaction monitoring model.
In the embodiment, time series characteristics of a series of factors such as medicines, symptoms and body index value changes in the clinical real world are brought into model learning, so that the model is stronger and more sensitive. Meanwhile, different elements are distributed and spliced according to the time sequence, so that the flexibility is higher, the incorporated different elements can be conveniently expanded, and the accuracy of model classification is further ensured. The adverse reactions of the current medication stage of the patient or the next medication stage after the medication of the patient is changed can be monitored and predicted, and whether the adverse reactions occur or which kind of adverse reactions occur can be accurately predicted based on the constructed model, so that corresponding preparation is made in advance, and the health of the patient is ensured.
Optionally, the acquiring medication information of the patient includes:
and acquiring the medication text.
Converting the medication text to a text vector based on a word space matrix.
And acquiring a time sequence feature vector, and fusing the text vector and the time sequence feature vector.
Illustratively, taking a large number of known adverse reaction sample texts as statistical objects, combing a uniform word space of an adverse reaction text corpus, and constructing a word space matrix of N x N, wherein N is the number of different words. Counting the frequency n of the common appearance of two characters from a large number of adverse reaction sample corpora through bigramijWherein n isijIndicating the frequency of the co-occurrence of the word in row i and the word in column j. A co-occurrence matrix between every two words is constructed according to the word space. From this matrix, words of the input text are converted into a vector representation.
As shown in table 1, is a uniform word space matrix of the adverse reaction text, wherein n is2565 indicates the co-occurrence frequency of the two words "skin" in row 2, column 5.
TABLE 1
Patient suffering from Leather Ground Plug for medical use Skin care products Western medicine Tide with water-collecting device Retreat Body of a shoe Red wine Mining Now that Adding And heavy load Super-super Speckle (mottle) ...
Patient suffering from 74 14 95 73 76 45 83 11 57 20 47 22 12 40 92 89 48
Leather 14 6 92 66 65 37 44 94 31 34 91 10 50 91 71 26 78
Ground 95 92 56 41 48 99 93 41 41 99 89 23 63 99 75 60 3
Plug for medical use 73 66 41 8 14 13 31 93 54 89 99 46 81 64 12 83 31
Skin care products 76 65 48 14 22 87 45 32 36 51 71 13 81 88 74 87 47
Western medicine 45 37 99 13 87 16 4 80 10 84 71 91 4 51 52 11 84
Tide with water-collecting device 83 44 93 31 45 4 6 38 18 33 23 28 58 96 59 17 97
Retreat 11 94 41 93 32 80 38 26 45 47 96 51 33 73 2 40 30
Body of a shoe 57 31 41 54 36 10 18 45 92 41 88 37 85 28 84 27 61
Red wine 20 34 99 89 51 84 33 47 41 14 33 39 60 75 51 21 73
Mining 47 91 89 99 71 71 23 96 88 33 18 20 18 31 31 59 12
Now that 22 10 23 46 13 91 28 51 37 39 20 57 36 78 25 31 82
Adding 12 50 63 81 81 4 58 33 85 60 18 36 29 1 28 10 17
And 40 91 99 64 88 51 96 73 28 75 31 78 1 88 55 98 69
heavy load 92 71 75 12 74 52 59 2 84 51 31 25 28 55 10 60 73
Super-super 89 26 60 83 87 11 17 40 27 21 59 31 10 98 60 96 94
Speckle (mottle) 48 78 3 31 47 84 97 30 61 73 12 82 17 69 73 94 97
...
Then the vector for the input text "skin flushing emphasis" can be selected from table 2 according to the uniform word space.
TABLE 2
Figure BDA0002733626390000061
Figure BDA0002733626390000071
Word vector s of corresponding word in each columniAnd i represents the ith word of the input, and the whole word forms a vector of the input text. Similarly, the overall vector S (S) of the medication text in which the patient information is mixed and spliced according to the time sequence can be obtained1,s2,.....,sn)。
Because the input medication texts are spliced in time series, from left to right, the earliest to the latest, the positions in the input texts correspond to the time series sequence. Let L be the length (i.e. number of words) of the input text, and the position corresponding to the ith word in the input text is piThen its corresponding timestamp value is:
Figure BDA0002733626390000072
the time series feature vector is ti＝(li,li,.....,li)，tiIs formed by d time stamp values liAnd d is the dimension of each word vector of the input text.
Will use medicine text vector S ═ S (S)1,s2,.....,sn) Fusing time series feature vectors, i.e. having
Figure BDA0002733626390000073
siE S, i.e. d dimensions of the word vector for each position are each added with a time tag value li。
Finally, a medication text vector fused with the time series feature vector can be obtained, which can be expressed as: x ═ X1,x2....xn)。
Optionally, the machine learning model comprises a recurrent neural network, an attention distribution model, a feed-forward network layer and a normalization function; the output of the recurrent neural network is used as the input of the attention degree distribution model, the output of the attention degree distribution model is used as the input of the feedforward network layer, and the output of the feedforward network layer is used as the input of the normalization function.
Optionally, the recurrent neural network is a bidirectional recurrent neural network, and a filtering mechanism is embedded in the bidirectional recurrent neural network.
In particular, a bidirectional Recurrent Neural Network (RNN) is used as an input feature extractor, and a filtering mechanism is embedded therein to control the manner in which information is updated. The forward propagation algorithm is as follows:
Figure BDA0002733626390000081
mt＝σ(Wmxt+Umft-1+bm)，mt∈[0,1]D；
Figure BDA0002733626390000082
εt＝σ(Wεxt+Uεft-1+bε)。
wherein x istRepresents the input, xte.X, σ (-) represents a nonlinear activation function, W, U and b are a network weight parameter and a bias respectively; f. oftIndicating the state at the current time;
Figure BDA0002733626390000083
representing a candidate state at the current time; m ist∈[0,1]DFor updating and filtering, the method is used for controlling how much information needs to be kept from the historical state in the current state; epsilont∈[0,1]DFor resetting filtering, for controlling candidate states
Figure BDA0002733626390000084
Whether or not to depend on the state f of the previous momentt-1。
The state output of the bidirectional recurrent neural network is:
Figure BDA0002733626390000085
the output sequence of the text with the length T passing through the bidirectional recurrent neural network is as follows: [ y ]1,y2,...yT]And T is the number of words of the input text.
Optionally, the output of the bidirectional recurrent neural network [ y ]1,y2,...yT]Will be used as an input sequence for the attention distribution model.
Let the input of the attention distribution model be X' ═ y1,y2,...yT]The output is U ═ U1,...un]. For each input yiThrough Tq，Tk，TvConverting the three matrixes to obtain an index vector qiTarget vector kiVector of sum values vi. The conversion process for the entire input X' is represented as follows:
Mq＝TqX
Mk＝TkX
Mv＝TvX′。
wherein, Tq，Tk，TvThree transformed parameter matrices. Mq＝[q1...qN]，Mk＝[k1...kN]，Mv＝[v1...vN]Respectively, a matrix composed of an index vector, a target vector, and a value vector.
For each index vector qnTarget vector kiVector of sum values viThe following output can be obtained from the attention distribution mechanism:
Figure BDA0002733626390000091
where n, i are the positions of the output and input vectors, aniIndicating that the ith input is output by the nth inputThe weight of interest. s (k)i,qn) For the attention evaluation function, d is the dimension of the target vector:
Figure BDA0002733626390000092
in this embodiment, a bidirectional recurrent neural network is used in combination with an attention distribution mechanism, so that the model can better learn the context semantic information to obtain the weight distribution of different positions of the input sequence. And because a plurality of groups of learning information are fused as model input, the attention degree distribution mechanism ensures that key information influencing final output can be effectively learned.
Optionally, the feed-forward network layer comprises:
F＝WFP+bF，F^＝ψ(F)；
wherein P represents an output of the attention distribution model, WFWeight of a parameter representing the feedforward network layer, bFRepresents the offset of the feedforward network layer, phi (-) represents the Logistic function, F represents the output of the feedforward network layer that has not been subjected to the activation function phi (-) and F (-) represents the output of the feedforward network layer after being subjected to the activation function phi (-).
Optionally, the output of the normalization function represents a two-class output with or without adverse reactions, or a multi-class output representing different classes of adverse reactions.
And the attention distribution model is connected with a fully-connected feedforward network layer and a softmax normalization function, the output of the attention distribution model is used as the input of the feedforward network layer, and finally the output softmax (F ^) of the softmax normalization function is used as the probability distribution of multi-classification. The classification can be two classifications with or without adverse reactions, or multiple classifications with different types of adverse reactions and without adverse reactions.
For example, if the output of the two-class classification is (0.96,0.04), the probability of having an adverse reaction is 0.96, and the probability of not having an adverse reaction is 0.04.
The output of the multi-classification is (0.1,0.15,0.7,0.05), and then different probabilities correspond to different adverse reaction categories respectively.
When the model is trained, for the two-classification task, each piece of data is labeled with a label with or without adverse reaction according to the known condition. For the multi-classification task, labeling the label corresponding to the adverse reaction category according to the known condition for each piece of data. And the two-classification task and the multi-classification task respectively use respective labeled data to train the model. Through training of a large amount of marked data, loss is calculated through a cross entropy loss function, and model parameters are obtained through iterative training and learning.
In another embodiment of the invention, an adverse reaction monitoring model construction device based on clinical real-world time series elements comprises a memory and a processor; the memory for storing a computer program; the processor is used for realizing the adverse reaction monitoring model construction method based on clinical real world time series elements when the computer program is executed.
As shown in fig. 2, an adverse reaction monitoring method based on clinical real-world time series elements according to an embodiment of the present invention includes the following steps:
s21, inputting the medication information of the calibrated patient into the adverse reaction monitoring model constructed by the adverse reaction monitoring model construction method based on the clinical real world time series elements, wherein the medication information is obtained based on a medication text and a time series characteristic vector, and the medication text comprises the information of the used medicines arranged according to the time sequence and corresponding symptom information and body index value change information.
S22, taking the output of the adverse drug reaction monitoring model as prediction data indicating the adverse reaction of the calibrated patient.
Specifically, in the practical application process, the acquired medication information of the calibration patients is collated according to the above-mentioned steps S11 and S12, and is used as the input of the trained model for prediction, so as to obtain probability distributions of different classifications, and the classification with the highest probability is used as the prediction answer.
In this embodiment, the method is suitable for the current medication stage of the patient, or scenes such as medication change, symptom sign change or body index value change. Such as the use of one drug at a time, followed by a change in medication for rescue or other purposes, thereby creating a need for monitoring. The time series characteristics of a series of factors such as medicines, symptoms, body index value changes and the like used in the clinical real world are brought into model learning, so that the model is stronger and more sensitive. Meanwhile, different elements are distributed and spliced according to the time sequence, so that the flexibility is higher, the incorporated different elements can be conveniently expanded, and the accuracy of model classification is further ensured. The adverse reactions of the current medication stage of the patient or the next medication stage after the medication of the patient is changed can be monitored and predicted, and whether the adverse reactions occur or which kind of adverse reactions occur can be accurately predicted based on the constructed model, so that corresponding preparation is made in advance, and the health of the patient is ensured.
In another embodiment of the present invention, an adverse reaction monitoring device based on clinical real-world time series elements comprises a memory and a processor; the memory for storing a computer program; the processor, when executing the computer program, is configured to implement the method for monitoring adverse reactions based on clinical real-world time series elements as described above.
In another embodiment of the present invention, a computer readable storage medium has stored thereon a computer program which, when executed by a processor, implements a method for constructing an adverse reaction monitoring model based on clinical real-world time series elements as described above, or implements a method for monitoring an adverse reaction based on clinical real-world time series elements as described above.
The reader should understand that in the description of this specification, reference to the description of the terms "one embodiment," "some embodiments," "an example," "a specific example" or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.

Claims (10)

1. An adverse reaction monitoring model construction method based on clinical real world time series elements is characterized by comprising the following steps:
acquiring medication information of a patient, wherein the medication information is obtained based on a medication text and a time series characteristic vector, and the medication text comprises used medicine information arranged according to a time sequence, corresponding symptom information and body index value change information;
acquiring adverse reaction information of the patient, wherein the adverse reaction information indicates whether the patient has an adverse reaction or the type of the occurred adverse reaction based on the medication information;
and training a preset machine learning model by using the medication information as a sample and the adverse reaction information as a label to obtain a medication adverse reaction monitoring model.
2. The method for constructing an adverse reaction monitoring model based on clinical real-world time series elements according to claim 1, wherein the obtaining of the medication information of the patient comprises:
acquiring the medication text;
converting the medication text into a text vector based on a word space matrix;
and acquiring a time sequence feature vector, and fusing the text vector and the time sequence feature vector.
3. The method for constructing an adverse reaction monitoring model based on clinical real-world time series elements according to claim 1, wherein the machine learning model comprises a recurrent neural network, an attention distribution model, a feedforward network layer and a normalization function; the output of the recurrent neural network is used as the input of the attention degree distribution model, the output of the attention degree distribution model is used as the input of the feedforward network layer, and the output of the feedforward network layer is used as the input of the normalization function.
4. The method for constructing the adverse reaction monitoring model based on the clinical real-world time series elements according to claim 3, wherein the recurrent neural network is a bidirectional recurrent neural network, and a filtering mechanism is embedded in the bidirectional recurrent neural network.
5. The method for constructing an adverse reaction monitoring model based on clinical real-world time series elements according to claim 3, wherein the feedforward network layer comprises:
F＝WFP+bF，F^＝ψ(F)；
wherein P represents an output of the attention distribution model, WFWeight of a parameter representing the feedforward network layer, bFRepresents the offset of the feedforward network layer, F represents the output of the feedforward network layer that has not been subjected to the activation function psi (·), and F ^ represents the output of the feedforward network layer after being subjected to the activation function psi (·).
6. The method for constructing an adverse reaction monitoring model based on clinical real-world time series elements according to claim 3, wherein the output of the normalization function represents two-classification output with or without adverse reaction or multi-classification output with different types of adverse reaction.
7. An adverse reaction monitoring model construction device based on clinical real world time series elements is characterized by comprising a memory and a processor; the memory for storing a computer program; the processor, when executing the computer program, implementing the method of any one of claims 1 to 6 for constructing a clinical real-world time series element-based adverse reaction monitoring model.
8. An adverse reaction monitoring method based on clinical real-world time series elements is characterized by comprising the following steps:
inputting the medication information of a calibrated patient into the adverse reaction monitoring model constructed by the adverse reaction monitoring model construction method based on clinical real world time series elements as claimed in any one of claims 1 to 6, wherein the medication information is obtained based on a medication text and a time series characteristic vector, and the medication text comprises the information of used medicines arranged according to time sequence and corresponding symptom information and body index value change information;
and taking the output of the adverse drug reaction monitoring model as prediction data for indicating the calibrated patient to generate adverse reaction.
9. An adverse reaction monitoring device based on clinical real world time series elements is characterized by comprising a memory and a processor; the memory for storing a computer program; the processor, when executing the computer program, for implementing the method of clinical real-world time series element based adverse reaction monitoring according to claim 8.
10. A computer-readable storage medium, characterized in that the storage medium has stored thereon a computer program which, when being executed by a processor, implements the method for constructing a clinical real-world time series element-based adverse reaction monitoring model according to any one of claims 1 to 6, or implements the method for monitoring an adverse reaction based on clinical real-world time series elements according to claim 8.
CN202011125974.4A 2020-10-20 2020-10-20 Adverse reaction monitoring method based on clinical real world time series elements Pending CN112366003A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011125974.4A CN112366003A (en) 2020-10-20 2020-10-20 Adverse reaction monitoring method based on clinical real world time series elements

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011125974.4A CN112366003A (en) 2020-10-20 2020-10-20 Adverse reaction monitoring method based on clinical real world time series elements

Publications (1)

Publication Number Publication Date
CN112366003A true CN112366003A (en) 2021-02-12

Family

ID=74506825

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011125974.4A Pending CN112366003A (en) 2020-10-20 2020-10-20 Adverse reaction monitoring method based on clinical real world time series elements

Country Status (1)

Country Link
CN (1) CN112366003A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113158073A (en) * 2021-03-25 2021-07-23 上海基绪康生物科技有限公司 Method for automatically discovering adverse drug reactions from social media
CN117558397A (en) * 2024-01-12 2024-02-13 吉林大学 Report generating system for analysis of deterioration condition of renal patients

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108376569A (en) * 2018-01-10 2018-08-07 中南大学湘雅三医院 A kind of drug induced disease monitoring method and system
CN109599189A (en) * 2019-01-11 2019-04-09 平安科技（深圳）有限公司 Method, apparatus, electronic equipment and the storage medium of medication abnormal response monitoring
CN110689939A (en) * 2019-08-27 2020-01-14 天津开心生活科技有限公司 Recommendation method and device for medication sequence, readable medium and electronic equipment
CN111063448A (en) * 2019-09-06 2020-04-24 中国人民解放军总医院 Establishment method, storage system and active early warning system of blood transfusion adverse reaction database
CN111383726A (en) * 2018-12-28 2020-07-07 国家食品药品监督管理总局药品评价中心 Electronic medical record data processing method and device, electronic equipment and readable medium

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108376569A (en) * 2018-01-10 2018-08-07 中南大学湘雅三医院 A kind of drug induced disease monitoring method and system
CN111383726A (en) * 2018-12-28 2020-07-07 国家食品药品监督管理总局药品评价中心 Electronic medical record data processing method and device, electronic equipment and readable medium
CN109599189A (en) * 2019-01-11 2019-04-09 平安科技（深圳）有限公司 Method, apparatus, electronic equipment and the storage medium of medication abnormal response monitoring
CN110689939A (en) * 2019-08-27 2020-01-14 天津开心生活科技有限公司 Recommendation method and device for medication sequence, readable medium and electronic equipment
CN111063448A (en) * 2019-09-06 2020-04-24 中国人民解放军总医院 Establishment method, storage system and active early warning system of blood transfusion adverse reaction database

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113158073A (en) * 2021-03-25 2021-07-23 上海基绪康生物科技有限公司 Method for automatically discovering adverse drug reactions from social media
CN117558397A (en) * 2024-01-12 2024-02-13 吉林大学 Report generating system for analysis of deterioration condition of renal patients
CN117558397B (en) * 2024-01-12 2024-03-12 吉林大学 Report generating system for analysis of deterioration condition of renal patients

Similar Documents

Publication Publication Date Title
Sidey-Gibbons et al. Machine learning in medicine: a practical introduction
CN109697285B (en) Hierarchical BilSt Chinese electronic medical record disease coding and labeling method for enhancing semantic representation
Che et al. Deep computational phenotyping
Bashir et al. An ensemble based decision support framework for intelligent heart disease diagnosis
CN109036577B (en) Diabetes complication analysis method and device
Che et al. Deep learning solutions to computational phenotyping in health care
CN111387938B (en) Patient heart failure death risk prediction system based on characteristic rearrangement one-dimensional convolutional neural network
Blanco et al. Multi-label clinical document classification: Impact of label-density
CN112366003A (en) Adverse reaction monitoring method based on clinical real world time series elements
Stylianou et al. EBM+: Advancing Evidence-Based Medicine via two level automatic identification of Populations, Interventions, Outcomes in medical literature
KR20210068713A (en) System for predicting disease progression using multiple medical data based on deep learning
Karaca et al. Comparison of deep learning and support vector machine learning for subgroups of multiple sclerosis
Albogamy et al. Decision support system for predicting survivability of hepatitis patients
Dudchenko et al. Extraction from Medical Records.
Zaghir et al. Real-world patient trajectory prediction from clinical notes using artificial neural networks and UMLS-based extraction of concepts
CN110633368A (en) Deep learning classification method for early colorectal cancer unstructured data
Lee et al. Linear classifier design in the weight space
Escudié et al. Deep representation for patient visits from electronic health records
Che et al. Time series feature learning with applications to health care
Rampogu A Review on the Use of Machine Learning Techniques in Monkeypox Disease Prediction
Naganure et al. Leveraging deep learning approaches for patient case similarity evaluation
Mohan Predicting post-procedural complications using neural networks on MIMIC-III data
Al-Ash et al. Indonesian protected health information removal using named entity recognition
CN117194604B (en) Intelligent medical patient inquiry corpus construction method
EP4105814A1 (en) Method and apparatus for segmenting a medical text report into sections

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
TA01 Transfer of patent application right

Effective date of registration: 20220311

Address after: 430000 C24, phase 3-1, accelerator, Wuhan biomedical park, No. 388, Gaoxin Second Road, Donghu New Technology Development Zone, Wuhan, Hubei Province

Applicant after: HUBEI HUADA GENE Research Institute

Address before: 430000, building 24, No. 388, Gaoxin 2nd Road, Donghu New Technology Development Zone, Jiangxia District, Wuhan City, Hubei Province

Applicant before: BGI WUHAN Co.,Ltd.

TA01 Transfer of patent application right