WO2020258483A1 - Clinical medication behavior analysis system based on highly effective negative sequential mining pattern, and working method therefor - Google Patents

Abstract

A clinical medication behavior analysis system based on a highly effective negative sequential mining pattern, and a working method therefor, the system comprising a data acquisition system and a behavior analysis system; the data acquisition system comprises a data acquisition module and a data transfer module; the data acquisition module, in real time, acquires and saves clinical medication behavior data; the data transfer module transfers the clinical medication behavior data to the behavior analysis system; the behavior analysis system comprises a data processing module, a data analysis module, and a data management module; the data processing module performs data cleaning and data classification on the clinical medication behavior data; the data analysis module performs analysis and predictions; and the data management module stores and displays analysis results and gives recommendations for the next step of medication. The present method applies an eNSP-IT algorithm to clinical medication behavior analysis, to rapidly identify the negative sequential relationship between medicines, better predict the next step of medication for the patient, and support clinical decisions based on a medication regimen change.

Description

A clinical medication behavior analysis system based on an efficient negative sequence mining model and its working method Technical field

The invention relates to a clinical medication behavior analysis system based on an efficient negative sequence mining mode and a working method thereof, and belongs to the technical field of application of negative sequence modes.

Background technique

In recent years, with the rapid development of my country's economy, people's physical fitness has been paid more and more attention, and medical care has also received more and more attention. With the continuous development of informatization, the medical information system has also made considerable progress in the process of transitioning from paper charts to electronic health records. At present, the medical information system has basically achieved electronic, digital and medialization. This transformation This has led to the accumulation of a large amount of data in the clinical data warehouse, and the medical industry has a large amount of data storage. These medical and health data include clinical diagnosis data, patient medication data, patient medical insurance data, and patient natural attribute information. How to discover valuable information, rules or knowledge, help doctors increase clinical knowledge, assist medical staff in diagnosis and treatment, and provide decision-making information for hospital managers has become a very socially valuable and urgent problem to be solved.

Data mining is the process of discovering hidden knowledge in a large information repository. Data mining techniques developed for retail or other industries can be applied to medical care. Data mining is a multidisciplinary research field that incorporates the latest theories and research methods such as database technology, pattern recognition, machine learning, fuzzy logic, artificial intelligence, information retrieval, statistics, high-performance computing, and neural networks.

As a frontier field of data mining, sequential pattern mining has attracted more and more scholars' attention. Sequential pattern mining refers to the mining of relatively time or other patterns with high frequency. It can discover potentially useful information and knowledge between transactions that people do not know in advance. In the field of health data analysis, one of the problems to be solved by sequential pattern analysis is that after the doctor prescribes a medicine to the patient, what kind of medicine will be used in a specific period of time in the future, and the difference between medicine and medicine, medicine and disease The process of interrelationship rules allows doctors to refer to past drug prescriptions when diagnosing and administering patients to accurately determine the patient’s next medication. Its main purpose is to study the sequence of clinical medications and find out the rules, that is, it is not only necessary to know whether the drug is used, but also to determine the order of use of the drug and other drugs, for example, a common gastritis treatment The order of medication is to prescribe glucose injection solution, prescribe vitamin 6, then prescribe cephalosporin injection, and finally prescribe sodium chloride injection. Therefore, the sequence mode can discover a frequent sequence in a certain period of time in the database, that is, which drugs will be used by doctors in this period of time, and the standard of more or less is determined by the minimum support. Each sequence is a group of combinations arranged according to the time of medication, and the minimum support can be set to mine sequences that meet different levels of frequency. However, when applying sequential patterns to analyze clinical medication behavior and predicting the next medication of patients, they only consider the events that have occurred, which is also called positive sequential pattern (PSP) mining.

With the deepening of research, researchers found that there is a lot of useful information hidden in non-occurring events, and this information is not available in pure positive sequence pattern mining, so relevant researchers began to mine negative sequence patterns ( Negative Sequential Pattern, NSP). Negative sequence mode involves not only events that have occurred, but also events that have not occurred. It can analyze and understand the potential meaning of the data more deeply, so as to dig out very valuable information that is easy to be ignored by people. For example: a, b, c, d,

Represents a medication sequence mode, which indicates that within a certain period of time, the patient used medication d after taking medication a and b without using medication c. Nowadays, the value of the negative sequence model is more and more recognized by people. It has an irreplaceable role in understanding and processing many medical applications, such as the analysis of patient medication behavior.

The patient medication record data in the hospital is the data source for mining. Take the diagnosis and treatment records of 5 patients within 2 months as an example, as shown in Table 1 is a transaction database sorted by patient ID and drug issuance time as keywords. A transaction database, a transaction represents a treatment situation, a single item represents the used medicine, and the letter in the single attribute records the medicine ID. Perform data preprocessing and organize the transaction database of Table 1 into the sequence database of Table 2.

Table 1

Table 2

Patient ID	The sequence of drugs used by the patient
1	{c}{i}
2	{a,b}{c}{a,d,f,g}
3	{c,e,g,h}
4	{c}{c,d,g,h}{i}
5	{i}

All medication records of a patient in a certain period of time constitute an ordered sequence, and the sequence is represented by <>. In the sequence, the items/item sets are in order, each item represents a kind of medicine, and the element refers to all medicines used by the patient at a specific point in time, denoted by {} or (), The patient may use the same Chinese medicine in different time periods, that is, an item may occur in different elements of a sequence. For example, the drug sequence with ID 2 in Table 2 is {a,b}{c}{a,d,f,g}. The patient used drug a during the first and third treatments, where {a ,b},{c},{a,d,f,g} these three itemsets can be called sequence elements, a,b,c,d,f,g are called items, if only one element For an item, the parentheses can be omitted. For example, the element {c} in the sequence can be directly written as c.

At present, there are few research results on negative sequence pattern mining algorithms, such as NSPM, PNSP, Neg-GSP, e-NSP and f-NSP, etc. However, most methods, even the most advanced algorithm f-NSP, are not efficient enough, and the number of negative sequence patterns mined is small. In practical applications, there are many factors that affect the efficiency and quantity of negative sequence pattern mining, the most important of which are the positive sequence pattern mining process and negative constraints. Since users are mainly interested in negative sequence patterns that lack some frequent elements, the existing negative sequence pattern mining algorithms first rely on identifying positive sequence patterns, but most algorithms ignore negative sequence patterns in the process of mining negative sequence patterns. In order to find the time consumption of the positive sequence pattern, this leads to a higher time cost of the entire mining process. At the same time, all negative sequence pattern algorithms restrict the format, frequency and negative elements from all aspects to reduce the number of negative candidate sequences and find specific negative sequence patterns of interest. To a certain extent, strict negative constraints can reduce the number of redundant negative candidate sequences and ensure computational efficiency, but will cause a lot of interesting negative sequence patterns to be lost, especially long-length negative sequence patterns (containing a lot of information) . In addition, in negative sequence pattern mining, negative constraint conditions will also affect the choice of negative candidate sequence generation method to a certain extent. When the constraint conditions change, the negative candidate sequence generation method should be changed accordingly.

Summary of the invention

Aiming at the shortcomings of the prior art, to increase the efficiency of mining negative sequences faster, and to discover more interesting negative sequence patterns, the present invention provides a clinical medication behavior analysis system based on efficient negative sequence mining patterns;

The invention also provides a working method of the above-mentioned clinical drug use behavior analysis system based on the efficient negative sequence mining mode.

The present invention proposes an efficient negative pattern mining algorithm named eNSP-IT. Applying the eNSP-IT algorithm to the analysis of clinical medication behavior can quickly find out the negative sequence relationship between medications, thereby better predicting the next medication of patients, and supporting clinical decision-making based on changes in medication regimens.

Term explanation:

1. Prefixspan algorithm: a classic positive sequence pattern mining algorithm, which is based on depth-first search. Its basic idea is to use frequent prefixes to divide the search space and projection sequence database, and search for related frequent sequences.

2. Database: Data set, referred to as DS, means a collection containing all data sequences.

3. Support: support, referred to as sup, indicates that the frequency of a candidate sequence in the database is called support.

4. Minimum support: minimum support, min_sup for short, indicates the minimum frequency of frequent patterns in the database, which is set by the user. When the support of the candidate sequence is greater than the minimum support, this candidate sequence is a frequent pattern.

5, a prefix, refers to two sequences assume _{α = <e 1 e 2 ...} e n> and _{β = <e 1 'e 2} ' ... e m '> (m≤n), if and only if e _{i' = e i (i≤m-1)} , e m '∈e m, and all (e _m -e _m' consecutive items in e _m ') are arranged in the alphabetical order, then β is α Of a prefix. In layman's terms, the prefix is the subsequence at the beginning of the sequence. For example, for the sequence B=<a(abc)(ac)d(cf)> and A=<a(abc)a>, then A is the prefix of B. Accordingly, the prefix for β, α is the projection of _{α '= <e m "e} m + 1 ... e n>, where _{e m" = (e m -e} m'). In layman's terms, projection refers to the largest subsequence of the sequence that does not contain a prefix. For example, the projection of sequence B relative to prefix A is B'=<cd(cf)>.

The technical scheme of the present invention is:

A clinical drug behavior analysis system based on an efficient negative sequence mining model, including a data acquisition system and a behavior analysis system that are connected through a transmission network communication;

The data acquisition system includes a data acquisition module and a data transmission module connected in sequence;

The data collection module is used to collect and save the patient's clinical medication behavior data in real time. The clinical medication behavior data includes the patient's ID number, timestamp (i.e., the time of diagnosis and treatment), prescribed drugs, symptoms, symptoms, and the department where the patient is located;

The data transmission module is used to transmit the patient's clinical medication behavior data to the behavior analysis system through a transmission network;

The behavior analysis system includes a data processing module, a data analysis module, and a data management module connected in sequence; and is arranged in a cloud server. The data transmission module is connected to the data processing module;

The data processing module is used to perform data cleaning on the collected clinical medication behavior data of the patient, and to classify the data according to the department and symptoms of the patient;

The data analysis module is used to analyze and predict the clinical medication behavior of the patient according to the processing result of the data processing module; the steps are as follows:

The data analysis module establishes the medication behavior sequence corresponding to the patient’s ID number based on the clinical medication behavior data processed by the data processing module, and combines the clinical medication behavior analysis method of the effective negative sequence mining mode to analyze the clinical medication behavior Behavior analysis and prediction. The clinical medication behavior data of patients in the department and the same symptoms constitute a sequence database. Each patient’s ID number corresponds to a patient’s medication records in a certain period of time to form an orderly sequence; Use an efficient negative sequence mining model of clinical drug behavior analysis method to mine the sequence database to obtain a negative sequence model that meets the minimum support requirements, that is, the commonly used treatment drugs for this disease, the order of medication, and the relationship between drugs and drugs. The negative sequence patterns that can be used for decision-making are screened out, and the patient's medication behavior is analyzed by using the sequence patterns for decision-making.

The data management module is used to store and display the processing results of the data processing module and the clinical medication behavior results analyzed by the data analysis module, and when the doctor prescribes the medication, the next medication is recommended. The data management module is used to view all clinical medication behavior records and all frequent clinical medication behaviors. When the doctor treats the patient, the system will provide the commonly used treatment plan for this disease, and when the first treatment plan is not satisfactory, it will provide an alternative treatment plan.

According to the present invention, preferably, the transmission network is a wired public network, a local area network or a 3G/4G network.

The invention adopts a cloud management platform design (such as Alibaba Cloud Server, Huawei Cloud, JD Cloud, etc.), and each hospital does not need to configure a server. The hospital rents the cloud management platform server of this system to help the hospital connect with the various system interfaces in the hospital and import data. You can log in to the system through the Internet at any place with the corresponding authority, without installing a client, and realize the flexibility of security management. This system can also be deployed in the hospital's local privatized cloud, and log in to the hospital's local area network to connect.

The working method of the above-mentioned clinical medication behavior analysis system based on the efficient negative sequence mining model includes the following steps:

(1) The data collection module collects and saves the patient's clinical medication behavior data in real time. The clinical medication behavior data includes the patient's ID number, timestamp (that is, the time of diagnosis and treatment), prescribed drugs, symptoms, symptoms, and the patient's department;

Set negative candidate sequence ns; for example, set a negative candidate sequence as

It means that the drugs b and d are not used, and a and c are the drugs a and b used;

Setting m-size refers to the m elements contained in the negative candidate sequence ns; for example,

Is a 4-size sequence;

Setting MPS(ns) refers to the largest positive subsequence of the negative candidate sequence ns, which is composed of all the positive elements contained in the negative candidate sequence ns in the original order; for example: in ns

Represents drugs not used, and a and c represent drugs used; the largest positive sequence is

Setting the positive pair P(ns) is the sequence after all the negative elements in a negative candidate sequence ns composed of the medicine used by the patient are converted into corresponding positive elements; for example,

Set 1-negMS _ns to refer to the subsequence of the negative candidate sequence ns, and the subsequence is composed of MPS(ns) and a negative element;

Setting 1-negMSS _ns refers to the set of subsequences of all negative sequences including the negative candidate sequence ns;

Setting p(1-negMS _ns ) means that the positive element in the sequence 1-negMS _ns remains unchanged, and the negative element is converted to the corresponding positive element; for example:

Setting ds refers to a data sequence in the database, ds contains the drugs used by a patient during this treatment, and the drugs are arranged in the order of medication;

In summary, for a data sequence ds and a sequence ns containing n negative elements with the number of all elements being m, satisfy the element constraint, format constraint, and frequent constraint, and meet the conditions:

And each 1-negMS _ns satisfies

Then ds contains ns:

Element constraint means: no negative items are allowed inside elements; only elements in the sequence can become negative; for example:

Meet the constraints; and

Does not meet the constraints because

Is the element

Internal negative

The format constraint means that there are no consecutive 2 or more negative elements; for example:

The constraint is not satisfied because the negative element

Are two consecutive negative elements;

Frequent constraint is: negative sequences satisfy 1-negMS _ns ∈1-negMSS _ns and _{p (1-negMS ns) ∈PSP} , PSP refers to the positive sequence pattern;

Frequent constraints consider the following aspects: (1) Users are interested in the lack of certain frequent elements in NSP. Therefore, the elements considered in NSP should have sufficient frequency. ENSP-IT requires that any p(1-negMS _ns ) belongs to PSP, which meets the requirement that every element in NSP is frequently present. (2) Users hope that NSP contains more useful information, which helps them make better decisions. (3) If we do not implement this constraint, the number of negative candidate sequences may be huge, or even unlimited, which will lead to very low NSP mining efficiency.

(2) The data transmission module transmits the patient's clinical medication behavior data to the behavior analysis system through the transmission network, and the behavior analysis system uses the eNSP-IT algorithm to analyze the clinical medication behavior data, including the following steps:

a. The data processing module performs data cleaning on the collected clinical medication behavior data of the patient, and classifies the data according to the department and disease of the patient;

b. The data analysis module analyzes and predicts the clinical medication behavior of the patient according to the processing result of the data processing module;

c. The data management module stores and displays the processing results of the data processing module and the clinical medication behavior results analyzed by the data analysis module, and when the doctor prescribes drugs, the next medication is recommended.

Preferably, according to the present invention, in step a, the data processing module performs data cleaning on the collected clinical medication behavior data of the patient, and classifies the data according to the patient's department and disease, including the following steps:

When collecting clinical medication behavior data of patients through the data collection system, a large amount of data will be generated, and at the same time there may be duplication in the data or incomplete data information. Therefore, need

d. Optimize the collected clinical medication behavior data of patients to make them suitable for later analysis. Data optimization includes filling in missing data and filtering out abnormal data;

e. Perform standardized processing on the optimized clinical medication behavior data of patients. The standardized processing refers to the integration of data, that is, the weekly medication records of patients with the same patient ID number are sorted into a sequence to form a complete sequence The clinical medication behavior data of patients; all medication records of a patient in a certain period of time constitute an ordered sequence. In the sequence, the item/item set is ordered, and each item represents a drug, and Elements refer to all medicines used by the patient at a specific point in time; the patient may use the same Chinese medicine in different time periods, that is, an item may occur in different elements in a sequence.

f. According to the two classification characteristics of the patient's department and disease, the patient's clinical drug behavior data is classified, and according to the patient's ID number, time stamp (that is, the time of diagnosis and treatment), drugs prescribed, symptoms, symptoms and the patient's department Stored in the data management module.

Preferably, according to the present invention, in step b, the data analysis module analyzes and predicts the patient's clinical medication behavior according to the processing result of the data processing module, including the following steps:

g. Use the modified positive sequence pattern mining algorithm Prefixspan to mine all positive sequence patterns, that is, the order of the most frequently used drugs in the patient population within a certain period of time. In the modified positive sequence pattern mining algorithm Prefixspan, right Every frequent positive sequence uses a bitmap to store the data sequence ID number that contains it;

h. The negative candidate sequence generation method of PNSP is used to generate negative candidate sequence (Negative Sequential Candidates, NSC), which is used to determine which drugs are used more frequently and which drugs are not used in a certain period of time ；

i. Use bitmap operations to calculate the support for negative candidate sequences;

j. Screen negative sequence patterns that meet the minimum support requirements from negative candidate sequences, and use appropriate screening methods to screen out negative sequence patterns that can be used for decision-making, and use the sequence patterns for decision-making on the patient's medication behavior Perform analysis; the doctor predicts the patient's next treatment plan based on the analysis result, and supports clinical decision-making based on changes in the drug plan. For example, two negative sequence patterns

with

P ₁ and P ₂ indicate that when treating gastritis, doctors often choose the prescriptions in these two sequences, and the potential relationship between the drugs in each prescription can be discovered through these two negative sequence patterns. P ₁ means that the doctor does not use vitamin C after using glucose, ceftriaxone, vitamin B6 and sodium chloride solution. P ₂ means that after the doctor prescribed ceftriaxone and vitamin C, he did not use vitamin C, and then used cimetidine instead of omeprazole. Therefore, using NSP mining methods can effectively help doctors accurately predict the patient's next medication.

According to the present invention, preferably, in step g, in order to improve the time efficiency of negative sequence pattern mining, the PrefixSpan algorithm is used to mine the positive sequence pattern, and at the same time, the Bitmap strategy is used to further enhance the PrefixSpan algorithm to improve space efficiency. Unlike other mining methods that use bitmap structures, the modified PrefixSpan algorithm uses simple bitmap structures and operations to obtain sequential patterns, including the following steps:

k. Add ID to each data sequence ds;

1. Scan the database (contains the collection of all data sequences ds) to find all items, the item refers to each medicine, create a bitmap for each item, the length of each bitmap is equal to the number of data sequences in the database, if one item If it appears in the data sequence i, the bitmap of the item is set to 1 at position i; otherwise, the bitmap of the item is set to 0 at position i, and the bitmap is represented by B; for example, the bitmap of item b is B (b)=|1|1|1|0|0|, it is included in the first, second and third data sequences.

m. Calculate the support of each item according to the bitmap of each item, that is, the number of 1 in the bitmap; determine whether the support of the item meets the minimum support min_sup, which is set by the user , The minimum frequency of frequent patterns; if the item's support is greater than or equal to the minimum support min_sup, then the item is a PSP of length 1, and the PSP of length 1 is regarded as a prefix of length 1; otherwise, it is not a length of 1 1 PSP, delete this item;

n. Perform recursive mining for each prefix of length i that meets the support requirements, i≥1, based on the bitmap of the prefix, find the data sequence containing the prefix, and store the projection of the data sequence corresponding to the prefix in the projection database ; For example, the bitmap of the prefix <a> is B(<a>)=|1|1|1|1|0|, which means it is contained in the first, second, third, and fourth data sequence , The projection database of the prefix <a> contains the projections of the first, second, third and fourth data series relative to the prefix <a> and the ID of the data series;

o. Scan the projection database to find all items, create a bitmap according to the ID of the corresponding data sequence, calculate the support of each item, that is, the number of 1 in the bitmap, if the support of all items is lower than min_sup, then Return recursively, otherwise, go to step p;

p. Combine the items that meet the support count and the current prefix, and perform bit operations on the bitmaps of the two, that is, perform the AND operation on the two bitmaps to obtain the new prefix and its bitmap. The new prefix is For a PSP with length i, if the PSP is a 1-size PSP, store its support directly, otherwise, continue to use the bitmap to store information;

q, i plus 1, the prefix is each new prefix after the merged item, and steps o to q are executed recursively.

According to the present invention, in step h, in order to increase the number of NSPs mined, ENSP-IT relaxes the frequent constraint, and at the same time adopts the PNSP negative candidate sequence generation method. The steps are as follows:

r. Generate 1-size NSC from 1-size PSP; such as 1-size PSP<a> generate 1-size

s. The definition constraint is: continuous negative elements in NSP are not allowed; 2-size NSC is generated by the arrangement of 1-size PSP and 1-size NSP, for example

If the last element of ns is a positive element, add 1-size PSP or 1-size NSP; otherwise, add 1-size PSP;

t. Add 1-size PSP or 1-size NSP to the (k-1)-size candidate sequence (NSC or PSP) to generate k-size NSC;

u. Repeat the above steps r to step t until no NSC is generated, or the number of NSC elements is greater than 2l+1, l represents the number of elements in the largest sequence in the PSP; if the number of elements in the largest sequence in the PSP is m, then generate The maximum number of elements of the NSP is 2m+1;

Further preferably, the k-size NSC is trimmed before calculating its support. The trimming method is:

in case

with

Then cut out the negative candidate sequence ns.

Further preferably, the k-size NSC is trimmed before calculating its support. The trimming method is:

in case

with

Then cut out the negative candidate sequence ns.

According to the present invention, the step i, calculating the support degree of the negative candidate sequence, refers to:

A sequence ns of size m and n negative elements, for

(Sequence containing only one negative element) ∈1-negMSS _ns (a set of sequences containing one negative element), 1≤i≤n, in the database, the support of ns sup(ns) is as in formula (I), (Ⅱ), formula (Ⅲ) shows:

If the size of ns is 1, and ns has only 1 negative element, the support of ns is:

If ns contains only one negative term, the support degree of sequence ns is:

sup(ns)=sup(MPS(ns)-sup(p(ns))) (Ⅱ)

Otherwise, the support of ns is:

In formula (Ⅰ), formula (Ⅱ), and formula (Ⅲ), OR refers to the AND operation in the bit operation, that is, the bitmap corresponding to p(1-negMS _i ) is ANDed one by one, and the AND operation means multiple The two bitmaps are merged to generate a new bitmap. If the same position in the bitmap is all 1, the corresponding position on the new bitmap is 1, otherwise, all are 0. N refers to the number of 1 in the bitmap. number. For example, a negative candidate sequence

sup<ce>=5, corresponding MPS(ns)=<ce>, p(1-negMS ₁ )=<ace>, p(1-negMS ₂ )=<cef>. Suppose B(<ace>)=|0|0|1|1|0|, B(<cef>)=|0|1|1|1|0|,

therefore

And

The beneficial effects of the present invention are:

1. At present, there are few algorithms for mining negative sequence patterns, but these algorithms are often very inefficient. We have proposed an efficient negative sequence pattern mining algorithm-eNSP-IT algorithm, which can mine in less time Sequence mode that users are interested in. For such data as clinical medication behavior data, dense data with many items and large sequence length have good experimental results, and the results can be obtained relatively quickly.

2. Compared with other negative sequence pattern mining algorithms, the negative constraint conditions of the eNSP-IT algorithm are more relaxed, which can mine more sequence patterns and provide users with more decision information.

3. The application of the present invention can fully combine the positive and negative sequence patterns as a reference in the process of clinical drug analysis, so as to discover the most commonly used drug treatment plan in the treatment of a certain disease, so that the doctor can treat the patient During treatment, the present invention can provide him with previous treatment plans, so as to better predict the patient's next medication and support clinical decision-making based on changes in the medication plan.

Description of the drawings

Fig. 1 is a structural block diagram of a clinical medication behavior analysis system based on an efficient negative sequence mining model of the present invention.

Detailed ways

In the following, the present invention is further limited in combination with the drawings and the embodiments of the specification, but is not limited thereto.

Example 1

A clinical drug behavior analysis system based on an efficient negative sequence mining model, as shown in Figure 1, includes a data acquisition system and a behavior analysis system connected through a transmission network communication;

The data acquisition system includes a data acquisition module and a data transmission module connected in sequence;

The data collection module is used to collect and save the patient's clinical medication behavior data in real time. The clinical medication behavior data includes the patient's ID number, timestamp (that is, the time of diagnosis and treatment), prescribed drugs, symptoms, symptoms, and the patient's department;

The data transmission module is used to transmit the patient's clinical medication behavior data to the behavior analysis system through the transmission network;

The behavior analysis system includes a data processing module, a data analysis module, and a data management module connected in sequence; and is set in the cloud server. The data transmission module is connected to the data processing module;

The data processing module is used to clean the collected clinical medication behavior data of the patient and classify the data according to the department and disease of the patient;

The data analysis module is used to analyze and predict the clinical medication behavior of patients according to the processing results of the data processing module; the steps are as follows:

The data analysis module establishes the medication behavior sequence corresponding to the patient’s ID number based on the clinical medication behavior data processed by the data processing module, and combines the clinical medication behavior analysis method of the effective negative sequence mining mode to analyze the clinical medication behavior Behavior analysis and prediction. The clinical medication behavior data of patients in the department and the same symptoms constitute a sequence database. Each patient’s ID number corresponds to a patient’s medication records in a certain period of time to form an orderly sequence; Use an efficient negative sequence mining model of clinical drug behavior analysis method to mine the sequence database to obtain a negative sequence model that meets the minimum support requirements, that is, the commonly used treatment drugs for this disease, the order of medication, and the relationship between drugs and drugs. The negative sequence patterns that can be used for decision-making are screened out, and the patient's medication behavior is analyzed by using the sequence patterns for decision-making.

The data management module is used to store and display the processing results of the data processing module and the clinical medication behavior results analyzed by the data analysis module. When the doctor prescribes drugs, the next medication is recommended. The data management module is used to view all clinical medication behavior records and all frequent clinical medication behaviors. When the doctor treats the patient, the system will provide the commonly used treatment plan for this disease, and when the first treatment plan is not satisfactory, it will provide an alternative treatment plan.

The transmission network is a wired public network, a local area network or a 3G/4G network.

The invention adopts a cloud management platform design (such as Alibaba Cloud Server, Huawei Cloud, JD Cloud, etc.), and each hospital does not need to configure a server. The hospital rents the cloud management platform server of this system to help the hospital connect with the various system interfaces in the hospital and import data. You can log in to the system through the Internet at any place through the corresponding authority, without installing a client, and realize the flexibility of security management. The system can also be deployed in the hospital's local privatized cloud, and log in to the hospital's local area network to connect.

Example 2

The working method of the clinical medication behavior analysis system based on the efficient negative sequence mining mode described in embodiment 1, includes the following steps:

(1) The data collection module collects and saves the patient's clinical medication behavior data in real time. The clinical medication behavior data includes the patient's ID number, timestamp (that is, the time of diagnosis and treatment), prescribed drugs, symptoms, symptoms, and the patient's department;

Set a negative candidate sequence ns composed of drugs used by the patient; for example, set a negative candidate sequence as

It means that the drugs b and d are not used, and a and c are the drugs a and b used;

Setting m-size refers to the m elements contained in the negative candidate sequence ns; for example,

Is a 4-size sequence;

Set MPS(ns) to be the largest positive subsequence of the negative candidate sequence ns composed of the medicine used by the patient, which is composed of all the positive elements contained in the negative candidate sequence ns in the original order, that is, this negative candidate sequence The composition of the drugs used by all patients in the

Represents drugs not used, and a and c represent drugs used; the largest positive sequence is

Setting the positive pair P(ns) is the sequence after all the negative elements in a negative candidate sequence ns composed of the medicine used by the patient are converted into corresponding positive elements; for example,

Set 1-negMS _ns to refer to the subsequence of the negative candidate sequence ns, and the subsequence is composed of MPS(ns) and a negative element;

Setting 1-negMSS _ns refers to the set of subsequences of all negative sequences including the negative candidate sequence ns;

Setting p(1-negMS _ns ) means that the positive element in the sequence 1-negMS _ns remains unchanged, and the negative element is converted to the corresponding positive element; for example:

Setting ds refers to a data sequence in the database, ds contains the drugs used by a patient during this treatment, and the drugs are arranged in the order of medication;

In summary, for a data sequence ds and a sequence ns containing n negative elements with the number of all elements being m, satisfy the element constraint, format constraint, and frequent constraint, and meet the conditions:

And each 1-negMS _ns satisfies

Then ds contains ns:

Element constraint means: no negative items are allowed inside elements; only elements in the sequence can become negative; for example:

Meet the constraints; and

Does not meet the constraints because

Is the element

Internal negative

The format constraint means that there are no consecutive 2 or more negative elements; for example:

The constraint is not satisfied because the negative element

Are two consecutive negative elements;

Frequent constraint is: negative sequences satisfy 1-negMS _ns ∈1-negMSS _ns and _{p (1-negMS ns) ∈PSP} , PSP refers to the positive sequence pattern;

(2) In this example, the gastritis outpatient data in the medical insurance data is used as the experimental data. Table 3 is a partial result of preprocessing the medical insurance data into a sequence database. The eNSP-IT algorithm is used to analyze the clinical medication behavior, and the minimum support Degree min_sup=30%, the data transmission module transmits the patient's clinical medication behavior data to the behavior analysis system through the transmission network. The behavior analysis system uses the eNSP-IT algorithm to analyze the clinical medication behavior data, including the following steps:

table 3

Patient ID	The sequence of drugs used by the patient
1	<(Glucose) (Sodium Chloride Solution) (Ceftriaxone) (Vitamin B6) (Cimetidine) (Domperidone)>
2	<(Omeprazole)(Amoxicillin)>
3	<(Sodium chloride solution)(Ceftriaxone)(Glucose)(Omeprazole)>
4	<(Sodium chloride solution)(Xiangdan injection)(Astragalus injection)>
5	<(Sodium Chloride Solution)(Ceftriaxone)(Diaoxinxuekang Capsule)(Sanjiuweitai Granule)(Domperidone)>
…	…

a. The data processing module cleans the collected clinical medication behavior data of the patient, and classifies the data according to the department and disease of the patient; the steps are as follows:

When collecting clinical medication behavior data of patients through the data collection system, a large amount of data will be generated, and at the same time there may be duplication in the data or incomplete data information. Therefore, need

d. Optimize the collected clinical medication behavior data of patients to make them suitable for later analysis. Data optimization includes filling in missing data and filtering out abnormal data;

e. Perform standardized processing on the optimized clinical medication behavior data of patients. The standardized processing refers to the integration of data, that is, the weekly medication records of patients with the same patient ID number are sorted into a sequence to form a complete sequence The clinical medication behavior data of patients; all medication records of a patient in a certain period of time constitute an ordered sequence. In the sequence, the item/item set is ordered, and each item represents a drug, and Elements refer to all medicines used by the patient at a specific point in time; the patient may use the same Chinese medicine in different time periods, that is, an item may occur in different elements in a sequence.

f. According to the two classification characteristics of the patient's department and disease, the patient's clinical drug behavior data is classified, and according to the patient's ID number, time stamp (that is, the time of diagnosis and treatment), drugs prescribed, symptoms, symptoms and the patient's department Stored in the data management module.

b. The data analysis module analyzes and predicts the clinical medication behavior of patients according to the processing results of the data processing module;

c. The data management module stores and displays the processing results of the data processing module and the clinical medication behavior results analyzed by the data analysis module. When the doctor prescribes drugs, the next medication is recommended.

Step b: The data analysis module analyzes and predicts the patient's clinical medication behavior according to the processing result of the data processing module, including the following steps:

g. Use the modified positive sequence pattern mining algorithm Prefixspan to mine all positive sequence patterns, that is, the order of the most frequently used drugs in the patient population within a certain period of time. In the modified positive sequence pattern mining algorithm Prefixspan, right Every frequent positive sequence uses a bitmap to store the data sequence ID number that contains it. Table 4 shows some positive sequence patterns and their bitmaps;

Table 4

Positive sequence mode	bitmap
<(Vitamin B6)(Vitamin C)>	|0|0|0|0|0|1|0|0|0|1|……|0|0|1|
<(Sodium Chloride Solution) (Ceftriaxone) (Omeprazole)>	|0|0|1|0|0|0|0|1|0|0|……|0|0|0|
<(Omeprazole) (Compound Daqing Leaf)>	|0|0|0|0|0|0|0|1|0|0|……|0|1|0|
<(Sanjiu Weitai Granules) (Domperidone)>	|0|0|0|0|1|0|0|0|1|0|……|1|0|0|
…	…

h. The negative candidate sequence generation method of PNSP is adopted to generate negative candidate sequence (Negative Sequential Candidates, NSC), which is used to determine which drugs are used more frequently and which drugs are not used in a certain period of time . According to the experimental data, generate the following negative candidate sequence

i. Use bitmap operations to calculate the support for negative candidate sequences;

j. Screen negative sequence patterns that meet the minimum support requirements from negative candidate sequences, and use appropriate screening methods to screen out negative sequence patterns that can be used for decision-making, and use the sequence patterns for decision-making on the patient's medication behavior Analyze; the doctor predicts the patient's next treatment plan according to the analysis result, and supports the clinical decision based on the change of the drug plan. Table 5 shows the partial negative sequence patterns mined under the minimum support min_sup=30%.

table 5

For example, two negative sequence patterns

with

P ₁ and P ₂ indicate that when treating gastritis, doctors often choose the prescriptions in these two sequences, and the potential relationship between the drugs in each prescription can be discovered through these two negative sequence patterns. P ₁ means that the doctor does not use vitamin C after using glucose, ceftriaxone, vitamin B6 and sodium chloride solution. P ₂ means that after the doctor prescribed ceftriaxone and vitamin C, he did not use vitamin C, and then used cimetidine instead of omeprazole. Therefore, using NSP mining methods can effectively help doctors accurately predict the patient's next medication.

According to the method described in step g, in order to improve the time efficiency of negative sequence pattern mining, the PrefixSpan algorithm is used to mine the positive sequence pattern. At the same time, the Bitmap strategy is used to further enhance the PrefixSpan algorithm to improve the space efficiency. Unlike other mining methods that use bitmap structures, the modified PrefixSpan algorithm uses simple bitmap structures and operations to obtain sequential patterns, including the following steps:

k. Add ID to each data sequence ds;

1. Scan the database (contains the collection of all data sequences ds) to find all items, the item refers to each medicine, create a bitmap for each item, the length of each bitmap is equal to the number of data sequences in the database, if one item If it appears in the data sequence i, the bitmap of the item is set to 1 at position i; otherwise, the bitmap of the item is set to 0 at position i, and the bitmap is represented by B;

m. Calculate the support of each item according to the bitmap of each item, that is, the number of 1 in the bitmap; determine whether the support of the item meets the minimum support min_sup, which is set by the user , The minimum frequency of frequent patterns; if the item's support is greater than or equal to the minimum support min_sup, then the item is a PSP of length 1, and the PSP of length 1 is regarded as a prefix of length 1; otherwise, it is not a length of 1 1 PSP, delete this item;

n. Perform recursive mining for each prefix of length i that meets the support requirements, i≥1, based on the bitmap of the prefix, find the data sequence containing the prefix, and store the projection of the data sequence corresponding to the prefix in the projection database ；

o. Scan the projection database to find all items, create a bitmap according to the ID of the corresponding data sequence, calculate the support of each item, that is, the number of 1 in the bitmap, if the support of all items is lower than min_sup, then Return recursively, otherwise, go to step p;

p. Combine the items that meet the support count and the current prefix, and perform bit operations on the bitmaps of the two, that is, perform the AND operation on the two bitmaps to obtain the new prefix and its bitmap. The new prefix is For a PSP with length i, if the PSP is a 1-size PSP, store its support directly, otherwise, continue to use the bitmap to store information;

q, i plus 1, the prefix is each new prefix after the merged item, and steps o to q are executed recursively.

According to the method described in step h, in order to increase the number of NSPs mined, ENSP-IT relaxes the frequent constraint and adopts the PNSP negative candidate sequence generation method. The steps are as follows:

r. Generate 1-size NSC from 1-size PSP; such as 1-size PSP<a> generate 1-size

s. The definition constraint is: continuous negative elements in NSP are not allowed; 2-size NSC is generated by the arrangement of 1-size PSP and 1-size NSP, for example

If the last element of ns is a positive element, add 1-size PSP or 1-size NSP; otherwise, add 1-size PSP;

t. Add 1-size PSP or 1-size NSP to the (k-1)-size candidate sequence (NSC or PSP) to generate k-size NSC;

u. Repeat the above steps r to step t until no NSC is generated, or the number of NSC elements is greater than 2l+1, l represents the number of elements in the largest sequence in the PSP; if the number of elements in the largest sequence in the PSP is m, then generate The maximum number of elements of the NSP is 2m+1;

The k-size NSC is trimmed before calculating its support. The trimming method is:

in case

with

Then cut out the negative candidate sequence ns.

The k-size NSC is trimmed before calculating its support. The trimming method is:

in case

with

Then cut out the negative candidate sequence ns.

According to the method described in step i, calculating the support degree of the negative candidate sequence refers to:

A sequence ns of size m and n negative elements, for

(Sequence containing only one negative element) ∈1-negMSS _ns (Set of sequences containing one negative element), 1≤i≤n, in the database, the support of ns sup(ns) is as formula (I), (Ⅱ), formula (Ⅲ) shows:

If the size of ns is 1, and ns has only 1 negative element, the support of ns is:

If ns contains only one negative term, the support degree of sequence ns is:

sup(ns)=sup(MPS(ns)-sup(p(ns))) (Ⅱ)

Otherwise, the support of ns is:

In formula (Ⅰ), formula (Ⅱ), and formula (Ⅲ), OR refers to the AND operation in the bit operation, that is, the bitmap corresponding to p(1-negMS _i ) is ANDed one by one, and the AND operation means multiple The two bitmaps are merged to generate a new bitmap. If the same position in the bitmap is all 1, the corresponding position on the new bitmap is 1, otherwise, all are 0. N refers to the number of 1 in the bitmap. number.

Example 2

The working method of the clinical medication behavior analysis system based on the efficient negative sequence mining mode described in embodiment 1, includes the following steps:

(1) The data collection module collects and saves the patient's clinical medication behavior data in real time. The clinical medication behavior data includes the patient's ID number, timestamp (that is, the time of diagnosis and treatment), prescribed drugs, symptoms, symptoms, and the patient's department;

Set a negative candidate sequence ns composed of drugs used by the patient; for example, set a negative candidate sequence as

It means that the drugs b and d are not used, and a and c are the drugs a and b used;

Setting m-size refers to the m elements contained in the negative candidate sequence ns; for example,

Is a 4-size sequence;

Set MPS(ns) to be the largest positive subsequence of the negative candidate sequence ns composed of the medicine used by the patient, which is composed of all the positive elements contained in the negative candidate sequence ns in the original order, that is, this negative candidate sequence The composition of the drugs used by all patients in the

Represents drugs not used, and a and c represent drugs used; the largest positive sequence is

Setting the positive pair P(ns) is the sequence after all the negative elements in a negative candidate sequence ns composed of the medicine used by the patient are converted into corresponding positive elements; for example,

Set 1-negMS _ns to refer to the subsequence of the negative candidate sequence ns, and the subsequence is composed of MPS(ns) and a negative element;

Setting 1-negMSS _ns refers to the set of subsequences of all negative sequences including the negative candidate sequence ns;

Setting p(1-negMS _ns ) means that the positive element in the sequence 1-negMS _ns remains unchanged, and the negative element is converted to the corresponding positive element; for example:

Setting ds refers to a data sequence in the database, ds contains the drugs used by a patient during this treatment, and the drugs are arranged in the order of medication;

In summary, for a data sequence ds and a sequence ns containing n negative elements with the number of all elements being m, satisfy the element constraint, format constraint, and frequent constraint, and meet the conditions:

And each 1-negMS _ns satisfies

Then ds contains ns:

Element constraint means: no negative items are allowed inside elements; only elements in the sequence can become negative; for example:

Meet the constraints; and

Does not meet the constraints because

Is the element

Internal negative

The format constraint means that there are no consecutive 2 or more negative elements; for example:

The constraint is not satisfied because the negative element

Are two consecutive negative elements;

Frequent constraint is: negative sequences satisfy 1-negMS _ns ∈1-negMSS _ns and _{p (1-negMS ns) ∈PSP} , PSP refers to the positive sequence pattern;

(2) In this embodiment, the data of diabetic patients in the medical insurance data is used as the experimental data. Table 6 below is the partial result of preprocessing the medical insurance data into a sequence database. The eNSP-IT algorithm is used to analyze the clinical medication behavior. Support min_sup=30%, including the following steps:

Table 6

Patient ID	The sequence of drugs used by the patient
1	<(Metformin, Simvastatin, Venlafaxine) (Aspirin, Glipizide) (Hydrochlorothiazide, Insulin)>

2	<(Metformin)(Glipizide)(Insulin)>
3	<(Aspirin, Azithromycin, Metformin) (Insulin)>
4	<(Metformin)(acetohexanamide)(rosiglitazone)>
5	<(Sulbutamide)(Metformin)(Alogliptin)(Piglitazone)(Exenatide)>
…	…

a. The data processing module cleans the collected clinical medication behavior data of the patient, and classifies the data according to the department and disease of the patient; the steps are as follows:

When collecting clinical medication behavior data of patients through the data collection system, a large amount of data will be generated, and at the same time there may be duplication in the data or incomplete data information. Therefore, need

d. Optimize the collected clinical medication behavior data of patients to make them suitable for later analysis. Data optimization includes filling in missing data and filtering out abnormal data;

e. Perform standardized processing on the optimized clinical medication behavior data of patients. The standardized processing refers to the integration of data, that is, the weekly medication records of patients with the same patient ID number are sorted into a sequence to form a complete sequence The clinical medication behavior data of patients; all medication records of a patient in a certain period of time constitute an ordered sequence. In the sequence, the item/item set is ordered, and each item represents a drug, and Elements refer to all medicines used by the patient at a specific point in time; the patient may use the same Chinese medicine in different time periods, that is, an item may occur in different elements in a sequence.

f. According to the two classification characteristics of the patient's department and disease, the patient's clinical drug behavior data is classified, and according to the patient's ID number, time stamp (that is, the time of diagnosis and treatment), drugs prescribed, symptoms, symptoms and the patient's department Stored in the data management module.

b. The data analysis module analyzes and predicts the clinical medication behavior of patients according to the processing results of the data processing module;

c. The data management module stores and displays the processing results of the data processing module and the clinical medication behavior results analyzed by the data analysis module. When the doctor prescribes drugs, the next medication is recommended.

Step b: The data analysis module analyzes and predicts the patient's clinical medication behavior according to the processing result of the data processing module, including the following steps:

g. Use the modified positive sequence pattern mining algorithm Prefixspan to mine all positive sequence patterns, that is, the order of the most frequently used drugs in the patient population within a certain period of time. In the modified positive sequence pattern mining algorithm Prefixspan, right Every frequent positive sequence uses a bitmap to store the data sequence ID number that contains it. Table 7 shows some positive sequence patterns and their bitmaps;

Table 7

Positive sequence mode	bitmap
<(Metformin)(Glipizide)>	|1|1|0|0|0|0|1|0|1|1|……|0|0|1|
<(Metformin)(Insulin)>	|1|1|1|0|0|0|1|0|0|0|……|1|0|0|
<(Glipizide) (Hydrochlorothiazide, Insulin)>	|1|0|0|0|0|1|0|0|0|0|……|0|0|0|
<(Aspirin)(Insulin)>	|1|0|1|0|0|0|0|0|1|0|……|1|0|0|
…	…

h. The negative candidate sequence generation method of PNSP is adopted to generate negative candidate sequence (Negative Sequential Candidates, NSC), which is used to determine which drugs are used more frequently and which drugs are not used in a certain period of time . According to the experimental data, generate the following negative candidate sequence

i. Use bitmap operations to calculate the support for negative candidate sequences;

j. Screen negative sequence patterns that meet the minimum support requirements from negative candidate sequences, and use appropriate screening methods to screen out negative sequence patterns that can be used for decision-making, and use the sequence patterns for decision-making on the patient's medication behavior Analyze; the doctor predicts the patient's next treatment plan based on the analysis result, and supports the clinical decision based on the change of the drug plan. Table 8 shows the partial negative sequence patterns mined under the minimum support min_sup=30%.

Table 8

For example, two negative sequence patterns

with

P ₁ and P ₂ show that when treating diabetes, doctors often choose prescriptions in these two sequences, and the potential relationship between the drugs in each prescription can be discovered through these two negative sequence patterns. P ₁ indicates that the doctor used metformin and not alogliptin after not using acetohexanamide. P ₂ means that after the doctor prescribed metformin, he did not use acetohexanamide and then used rosiglitazone instead of saxagliptin. Therefore, using NSP mining methods can effectively help doctors accurately predict the patient's next medication.

According to the method described in step g, in order to improve the time efficiency of negative sequence pattern mining, the PrefixSpan algorithm is used to mine the positive sequence pattern. At the same time, the Bitmap strategy is used to further enhance the PrefixSpan algorithm to improve the space efficiency. Unlike other mining methods that use bitmap structures, the modified PrefixSpan algorithm uses simple bitmap structures and operations to obtain sequential patterns, including the following steps:

k. Add ID to each data sequence ds;

1. Scan the database (contains the collection of all data sequences ds) to find all items, the item refers to each medicine, create a bitmap for each item, the length of each bitmap is equal to the number of data sequences in the database, if one item If it appears in the data sequence i, the bitmap of the item is set to 1 at position i; otherwise, the bitmap of the item is set to 0 at position i, and the bitmap is represented by B; for example, the item of sodium chloride solution The bitmap is B(b)=|1|1|1|0|0|, which is included in the first, second and third data sequences.

m. Calculate the support of each item according to the bitmap of each item, that is, the number of 1 in the bitmap; determine whether the support of the item meets the minimum support min_sup, which is set by the user , The minimum frequency of frequent patterns; if the item's support is greater than or equal to the minimum support min_sup, then the item is a PSP of length 1, and the PSP of length 1 is regarded as a prefix of length 1; otherwise, it is not a length of 1 1 PSP, delete this item;

n. Perform recursive mining for each prefix of length i that meets the support requirements, i≥1, based on the bitmap of the prefix, find the data sequence containing the prefix, and store the projection of the data sequence corresponding to the prefix in the projection database ; For example, the bitmap of the prefix <a> is B(<a>)=|1|1|1|1|0|, which means it is contained in the first, second, third, and fourth data sequence , The projection database of the prefix <a> contains the projections of the first, second, third and fourth data series relative to the prefix <a> and the ID of the data series;

o. Scan the projection database to find all items, create a bitmap according to the ID of the corresponding data sequence, calculate the support of each item, that is, the number of 1 in the bitmap, if the support of all items is lower than min_sup, then Return recursively, otherwise, go to step p;

p. Combine the items that meet the support count and the current prefix, and perform bit operations on the bitmaps of the two, that is, perform the AND operation on the two bitmaps to obtain the new prefix and its bitmap. The new prefix is For a PSP with length i, if the PSP is a 1-size PSP, store its support directly, otherwise, continue to use the bitmap to store information;

q, i plus 1, the prefix is each new prefix after the merged item, and steps o to q are executed recursively.

According to the method described in step h, in order to increase the number of NSPs mined, ENSP-IT relaxes the frequent constraint and adopts the PNSP negative candidate sequence generation method. The steps are as follows:

r. Generate 1-size NSC from 1-size PSP; such as 1-size PSP<a> generate 1-size

s. The definition constraint is: continuous negative elements in NSP are not allowed; 2-size NSC is generated by the arrangement of 1-size PSP and 1-size NSP, for example

If the last element of ns is a positive element, add 1-size PSP or 1-size NSP; otherwise, add 1-size PSP;

t. Add 1-size PSP or 1-size NSP to the (k-1)-size candidate sequence (NSC or PSP) to generate k-size NSC;

u. Repeat the above steps 1 to n until no NSC is generated, or the number of NSC elements is greater than 2l+1, l represents the number of elements in the largest sequence in the PSP; if the number of elements in the largest sequence in the PSP is m, then generate The maximum number of elements of the NSP is 2m+1;

The k-size NSC is trimmed before calculating its support. The trimming method is:

in case

with

Then cut out the negative candidate sequence ns.

The k-size NSC is trimmed before calculating its support. The trimming method is:

in case

with

Then cut out the negative candidate sequence ns.

According to the method described in step i, calculating the support degree of the negative candidate sequence refers to:

A sequence ns of size m and n negative elements, for

(Sequence containing only one negative element) ∈1-negMSS _ns (Set of sequences containing one negative element), 1≤i≤n, in the database, the support of ns sup(ns) is as formula (I), (Ⅱ), formula (Ⅲ) shows:

If the size of ns is 1, and ns has only 1 negative element, the support of ns is:

If ns contains only one negative term, the support degree of sequence ns is:

sup(ns)=sup(MPS(ns)-sup(p(ns))) (Ⅱ)

Otherwise, the support of ns is:

In formula (Ⅰ), formula (Ⅱ), and formula (Ⅲ), OR refers to the AND operation in the bit operation, that is, the bitmap corresponding to p(1-negMS _i ) is ANDed one by one, and the AND operation means multiple The two bitmaps are merged to generate a new bitmap. If the same position in the bitmap is all 1, the corresponding position on the new bitmap is 1, otherwise, all are 0. N refers to the number of 1 in the bitmap. number. For example, a negative candidate sequence

sup<ce>=5, corresponding MPS(ns)=<ce>, p(1-negMS ₁ )=<ace>, p(1-negMS ₂ )=<cef>. Suppose B(<ace>)=|0|0|1|1|0|, B(<cef>)=|0|1|1|1|0|,

therefore

And

Algorithm pseudo code

Input: clinical medication record sequence database (D); minimum support (min_sup);

Output: Sequence pattern set (NSP) used to analyze clinical medication behavior;

Step (1) is to use the modified PrefixSpan algorithm to dig out all positive sequence patterns from the sequence database, and the support of all positive candidate sequences are stored using bitmaps;

Steps (2)-(19) refer to generating negative candidates using a negative candidate sequence generation method, where steps (10) and (16) represent pruning the negative candidate sequences that meet the pruning conditions;

Steps (21)-(26) means using formulas (Ⅰ)-(Ⅲ) to calculate the support of negative candidate sequences, where steps (21)-(24) refer to calculating the support of negative candidates containing only one negative element. Step (26) refers to calculating the support degree of negative candidates containing multiple negative elements;

Steps (27)-(28) means that if the support of the negative candidate is greater than the minimum support, then this negative candidate sequence is a negative sequence pattern and is added to the set of negative sequence patterns

Step (30) refers to returning the results, and then using appropriate methods to screen out the sequence patterns that can be used for decision-making, and use these screened sequence patterns to analyze the clinical medication behavior.

Claims (10)

1. A clinical medication behavior analysis system based on an efficient negative sequence mining model, which is characterized in that it includes a data acquisition system and a behavior analysis system that are communicatively connected through a transmission network;

   The data collection system includes a data collection module and a data transmission module that are sequentially connected; the data collection module is used to collect and save the patient's clinical medication behavior data in real time. The clinical medication behavior data includes the patient's ID number, time stamp, and issuance The data transmission module is used to transmit the clinical medication behavior data of the patient to the behavior analysis system through the transmission network;

   The behavior analysis system includes a data processing module, a data analysis module, and a data management module that are sequentially connected; the data processing module is used to perform data cleaning on the collected clinical medication behavior data of the patient, and perform data processing according to the department and disease of the patient Data classification; the data analysis module is used to analyze and predict the clinical medication behavior of patients according to the processing results of the data processing module; the data management module is used to analyze and predict the processing results and data of the data processing module The clinical medication behavior results analyzed by the analysis module are stored and displayed. When the doctor prescribes the medication, the next medication is recommended.
2. The clinical medication behavior analysis system based on an efficient negative sequence mining model according to claim 1, wherein the transmission network is a wired public network, a local area network, or a 3G/4G network.
3. The working method of a clinical medication behavior analysis system based on an efficient negative sequence mining model according to claim 1 or 2, characterized in that it comprises the following steps:

   (1) The data collection module collects and saves the patient's clinical medication behavior data in real time. The clinical medication behavior data includes the patient's ID number, timestamp, prescribed drugs, symptoms, symptoms, and the patient's department;

   Set negative candidate sequence ns;

   Setting m-size refers to the m elements contained in the negative candidate sequence ns;

   Set MPS(ns) to refer to the largest positive subsequence of the negative candidate sequence ns, which is composed of all the positive elements contained in the negative candidate sequence ns in the original order;

   Setting the positive pair P(ns) is the sequence after all the negative elements in a negative candidate sequence ns composed of the medicine used by the patient are converted into the corresponding positive elements;

   Set 1-negMS _ns to refer to the subsequence of the negative candidate sequence ns, and the subsequence is composed of MPS(ns) and a negative element;

   Setting 1-negMSS _ns refers to the set of subsequences of all negative sequences including the negative candidate sequence ns;

   Setting p(1-negMS _ns ) means that the positive element in the sequence 1-negMS _ns remains unchanged, and the negative element is converted to the corresponding positive element;

   Setting ds refers to a data sequence in the database, ds contains the drugs used by a patient during this treatment, and the drugs are arranged in the order of medication;

   In summary, for a data sequence ds and a sequence ns containing n negative elements with the number of all elements being m, satisfy the element constraint, format constraint, and frequent constraint, and meet the conditions:

   

   And each 1-negMS _ns satisfies

   

   Then ds contains ns:

   Element constraint means: no negative items are allowed inside the element; only elements in the sequence can become negative;

   The format constraint means: there are no consecutive 2 or more negative elements;

   Frequent constraint is: negative sequences satisfy 1-negMS _ns ∈1-negMSS _ns and _{p (1-negMS ns) ∈PSP} , PSP refers to the positive sequence pattern;

   (2) The data transmission module transmits the patient's clinical medication behavior data to the behavior analysis system through the transmission network, and the behavior analysis system uses the eNSP-IT algorithm to analyze the clinical medication behavior data, including the following steps:

   a. The data processing module performs data cleaning on the collected clinical medication behavior data of the patient, and classifies the data according to the department and disease of the patient;

   b. The data analysis module analyzes and predicts the clinical medication behavior of the patient according to the processing result of the data processing module;

   c. The data management module stores and displays the processing results of the data processing module and the clinical medication behavior results analyzed by the data analysis module, and when the doctor prescribes drugs, the next medication is recommended.
4. The working method of a clinical medication behavior analysis system based on an efficient negative sequence mining model according to claim 3, wherein in step a, the data processing module performs data processing on the collected clinical medication behavior data of patients Data cleaning, and data classification according to the department and disease of the patient, including the following steps:

   d. Optimize the collected clinical medication behavior data of patients, and optimize the data including filling in missing data and filtering out abnormal data;

   e. Perform standardized processing on the optimized clinical medication behavior data of patients. The standardized processing refers to the integration of data, that is, the weekly medication records of patients with the same patient ID number are sorted into a sequence to form a complete sequence Clinical medication behavior data of patients;

   f. According to the two classification characteristics of the patient's department and disease, the clinical drug behavior data of the patient is classified, and the data management is stored in the data management according to the patient's ID number, timestamp, drugs prescribed, symptoms, symptoms and the patient's department Module.
5. The working method of a clinical drug use behavior analysis system based on an efficient negative sequence mining model according to claim 3, wherein, in step b, the data analysis module performs an analysis on the basis of the processing result of the data processing module The analysis and prediction of the patient's clinical medication behavior includes the following steps:

   g. Use the modified positive sequence pattern mining algorithm Prefixspan to mine all positive sequence patterns, that is, the order of the most frequently used drugs in the patient population within a certain period of time. In the modified positive sequence pattern mining algorithm Prefixspan, right Every frequent positive sequence uses a bitmap to store the data sequence ID number that contains it;

   h. The negative candidate sequence generation method of PNSP is used to generate negative candidate sequence (Negative Sequential Candidates, NSC), which is used to determine which drugs are used more frequently and which drugs are not used in a certain period of time ；

   i. Use bitmap operations to calculate the support for negative candidate sequences;

   j. Screen negative sequence patterns that meet the minimum support requirements from negative candidate sequences, and use appropriate screening methods to screen out negative sequence patterns that can be used for decision-making, and use the sequence patterns for decision-making on the patient's medication behavior Perform analysis; the doctor predicts the patient's next treatment plan based on the analysis result, and supports clinical decision-making based on changes in the drug plan.
6. The working method of a clinical medication behavior analysis system based on an efficient negative sequence mining model according to claim 5, wherein the step g includes the following steps:

   k. Add ID to each data sequence ds;

   1. Scan the database to find all items. The item refers to each drug. Create a bitmap for each item. The length of each bitmap is equal to the number of data sequences in the database. If an item appears in the data sequence i, then The bitmap of the item is set to 1 at position i; otherwise, the bitmap of the item is set to 0 at position i, and the bitmap is represented by B;

   m. Calculate the support of each item according to the bitmap of each item, that is, the number of 1 in the bitmap; determine whether the support of the item meets the minimum support min_sup, which is set by the user , The minimum frequency of frequent patterns; if the item's support is greater than or equal to the minimum support min_sup, then the item is a PSP of length 1, and the PSP of length 1 is regarded as a prefix of length 1; otherwise, it is not a length of 1 1 PSP, delete this item;

   n. Perform recursive mining for each prefix of length i that meets the support requirements, i≥1, based on the bitmap of the prefix, find the data sequence containing the prefix, and store the projection of the data sequence corresponding to the prefix in the projection database ；

   o. Scan the projection database to find all items, create a bitmap according to the ID of the corresponding data sequence, calculate the support of each item, that is, the number of 1 in the bitmap, if the support of all items is lower than min_sup, then Return recursively, otherwise, go to step p;

   p. Combine the items that meet the support count and the current prefix, and perform bit operations on the bitmaps of the two, that is, perform the AND operation on the two bitmaps to obtain the new prefix and its bitmap. The new prefix is For a PSP with length i, if the PSP is a 1-size PSP, store its support directly, otherwise, continue to use the bitmap to store information;

   q, i plus 1, the prefix is each new prefix after the merged item, and steps o to q are executed recursively.
7. The working method of a clinical medication behavior analysis system based on an efficient negative sequence mining model according to claim 5, wherein the step h includes the following steps:

   r. Generate 1-size NSC from 1-size PSP;

   s. The definition constraint is: continuous negative elements in NSP are not allowed; 2-size NSC is generated by the arrangement of 1-size PSP and 1-size NSP, if the last element of ns is a positive element, then Attach 1-size PSP or 1-size NSP; otherwise, attach 1-size PSP;

   t. Add 1-size PSP or 1-size NSP to the (k-1)-size candidate sequence to generate k-size NSC;

   u. Repeat the above steps r to step t until no NSC is generated, or the number of NSC elements is greater than 2l+1, l represents the number of elements in the largest sequence in the PSP; if the number of elements in the largest sequence in the PSP is m, then generate The maximum number of elements of the NSP is 2m+1.
8. The working method of a clinical medication behavior analysis system based on an efficient negative sequence mining model according to claim 5, wherein the k-size NSC is pruned before calculating its support, and the pruning method is:

   in case

   

   with

   

   Then cut out the negative candidate sequence ns.
9. The working method of a clinical medication behavior analysis system based on an efficient negative sequence mining model according to claim 5, wherein the k-size NSC is pruned before calculating its support, and the pruning method is:

   in case

   

   with

   

   Then cut out the negative candidate sequence ns.
10. The working method of a clinical medication behavior analysis system based on an efficient negative sequence mining model according to any one of claims 5-9, wherein the step i, calculating the support degree of the negative candidate sequence, refers to:

    A sequence ns of size m and n negative elements, for

    

    (Sequence containing only one negative element) ∈1-negMSS _ns (Set of sequences containing one negative element), 1≤i≤n, in the database, the support of ns sup(ns) is as formula (I), (Ⅱ), formula (Ⅲ) shows:

    If the size of ns is 1, and ns has only 1 negative element, the support of ns is:

    

    If ns contains only one negative term, the support degree of sequence ns is:

    sup(ns)=sup(MPS(ns)-sup(p(ns))) (Ⅱ)

    Otherwise, the support of ns is:

    

    In formula (Ⅰ), formula (Ⅱ), and formula (Ⅲ), OR refers to the AND operation in the bit operation, that is, the bitmap corresponding to p(1-negMS _i ) is ANDed one by one, and the AND operation means multiple The two bitmaps are merged to generate a new bitmap. If the same position in the bitmap is all 1, the corresponding position on the new bitmap is 1, otherwise, all are 0. N refers to the number of 1 in the bitmap. number.

Images