Method, device and storage medium for identifying abnormal medical records

Abstract

The embodiments of the present application provide a method, device and storage medium for identifying abnormal medical records. Taking the patient as the identification unit, based on the diagnosis set and medical item set contained in the patient's medical records within a preset time, the diagnosis set sequence and medical item set sequence corresponding to the patient are constructed according to the medical time; based on these two set sequences, the diagnosis set correlation and medical item set correlation between two adjacent medical visits are calculated in turn to generate a diagnosis set correlation sequence and a medical item set correlation sequence; these two correlation sequences can be used as the basis for identifying abnormal medical records. If these two correlation sequences do not meet the preset conditions, it can be determined that the medical records of the current patient are abnormal. In this way, based on the aforementioned diagnosis set correlation sequence and medical item set correlation sequence, it can be perceived whether the correlation between the patient's multiple medical visits is abnormal, thereby efficiently / accurately realizing the identification of abnormal medical records.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device, and storage medium for identifying abnormal medical records. Background Art

[0002] To strengthen the supervision and management of the use of medical insurance funds, ensure fund security, promote their effective use, and safeguard citizens' legitimate medical insurance rights and interests, official agencies have formulated and promulgated the "Regulations on the Supervision and Management of the Use of Medical Insurance Funds." Medical insurance funds are used by numerous entities, with a long supply chain and numerous risk points, making supervision difficult. This has led to frequent violations and fraud.

[0003] Among various forms of fraud and insurance deception, borrowing medical insurance cards and falsifying medical records are the most common, so such behaviors need to be identified and supervised. Summary of the Invention

[0004] Various aspects of the present application provide a method, device, and storage medium for identifying abnormal medical records, so as to more efficiently and accurately identify abnormal medical records.

[0005] This embodiment of the present application provides a method for identifying abnormal medical records, including:

[0006] Extract diagnosis sets and medical item sets from the target patient's medical records within a preset time period, with each medical visit corresponding to one medical record.

[0007] According to the time of the consultation, the diagnosis set and medical item set occurring in each consultation are sorted respectively to obtain the diagnosis set sequence and the medical item set sequence;

[0008] Based on the diagnosis set sequence, calculating the diagnosis set correlation between any two adjacent visits to generate a diagnosis set correlation sequence;

[0009] Based on the medical item set sequence, calculating the medical item set correlation between any two adjacent medical visits to generate a medical item set correlation sequence;

[0010] If the diagnosis set correlation sequence and the medical item set correlation sequence do not meet the preset conditions, it is determined that there is an abnormality in the medical record of the target patient.

[0011] An embodiment of the present application further provides a computing device, including a memory and a processor;

[0012] The memory is used to store one or more computer instructions;

[0013] The processor is coupled to the memory and configured to execute the one or more computer instructions for:

[0014] Extract diagnosis sets and medical item sets from the target patient's medical records within a preset time period, with each medical visit corresponding to one medical record.

[0015] According to the time of the consultation, the diagnosis set and medical item set occurring in each consultation are sorted respectively to obtain the diagnosis set sequence and the medical item set sequence;

[0016] Based on the diagnosis set sequence, calculating the diagnosis set correlation between any two adjacent visits to generate a diagnosis set correlation sequence;

[0017] Based on the medical item set sequence, calculating the medical item set correlation between any two adjacent medical visits to generate a medical item set correlation sequence;

[0018] If the diagnosis set correlation sequence and the medical item set correlation sequence do not meet the preset conditions, it is determined that there is an abnormality in the medical record of the target patient.

[0019] An embodiment of the present application also provides a computer-readable storage medium storing computer instructions. When the computer instructions are executed by one or more processors, the one or more processors are caused to execute the aforementioned method for identifying abnormal medical records.

[0020] In an embodiment of the present application, the patient is used as the identification unit, and based on the diagnosis set and medical item set contained in the patient's medical records within a preset time, the diagnosis set sequence and medical item set sequence corresponding to the patient are constructed according to the medical time; the diagnosis set correlation and medical item set correlation between two adjacent medical visits are calculated in sequence within the diagnosis set sequence and the medical item set sequence respectively to generate a diagnosis set correlation sequence and a medical item set correlation sequence; these two correlation sequences can be used as the basis for identifying abnormal medical records. If these two correlation sequences do not meet the preset conditions, it can be determined that the medical records of the current patient are abnormal. In this way, based on the aforementioned diagnosis set correlation sequence and medical item set correlation sequence, it is possible to efficiently / accurately perceive whether there is an abnormal correlation between multiple medical visits of the patient, thereby efficiently / accurately realizing the identification of abnormal medical records. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0022] Figure 1 A flowchart of a method for identifying abnormal medical records provided by an exemplary embodiment of the present application;

[0023] Figure 2a A logical diagram of a method for identifying abnormal medical records provided by an exemplary embodiment of the present application;

[0024] Figure 2b A logical diagram of an optional implementation method provided for an exemplary embodiment of the present application;

[0025] Figure 3 A schematic diagram of the analysis results corresponding to the correlation sequences of several patients provided in an exemplary embodiment of the present application;

[0026] Figure 4 A schematic diagram of a calculation scheme for the relevance of a medical project set provided by an exemplary embodiment of the present application;

[0027] Figure 5 A schematic diagram of an implementation method for calculating the correlation between medical project sets provided by an exemplary embodiment of the present application;

[0028] Figure 6 A schematic structural diagram of a computing device provided as another exemplary embodiment of the present application. DETAILED DESCRIPTION

[0029] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0030] Currently, it is usually necessary to identify abnormal medical records based on a large number of rigid rules, which is inefficient and has insufficient accuracy and recall. To this end, in some embodiments of the present application: the inventors found that under current medical insurance fraud methods such as borrowing medical insurance cards and falsifying medical records, the medical item records purchased and consumed for patients are actually not real needs or have not actually occurred. Based on this, it is proposed to use patients as identification units, and based on the diagnosis sets and medical item sets contained in the patient's medical records within a preset time, construct a diagnosis set sequence and a medical item set sequence corresponding to the patient according to the medical visit time; within the diagnosis set sequence and the medical item set sequence, the diagnosis set correlation and the medical item set correlation between two adjacent medical visits are calculated respectively to generate a diagnosis set correlation sequence and a medical item set correlation sequence; these two correlation sequences can be used as the basis for identifying abnormal medical records. If these two correlation sequences do not meet the preset conditions, it can be determined that the current patient's medical records are abnormal. In this way, based on the aforementioned diagnosis set correlation sequence and medical item set correlation sequence, it is possible to efficiently and accurately perceive whether there is an abnormal correlation between multiple medical visits of a patient, thereby efficiently and accurately achieving the identification of abnormal medical records.

[0031] The technical solutions provided by the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0032] Figure 1 A flowchart of a method for identifying abnormal medical records provided by an exemplary embodiment of the present application is provided. Figure 2a A logical diagram of a method for identifying abnormal medical records provided by an exemplary embodiment of the present application. The method can be executed by a data processing device, which can be implemented as a combination of software and / or hardware, and the data processing device can be integrated into a computing device. Figure 1 , the method may include:

[0033] Step 100: Extract a diagnosis set and a medical item set from the target patient's medical records within a preset time period. Each medical visit corresponds to one medical record.

[0034] Step 101: Sort the diagnosis set and medical item set generated in each visit according to the visit time to obtain a diagnosis set sequence and a medical item set sequence;

[0035] Step 102: Calculate the diagnosis set correlation between any two adjacent visits based on the diagnosis set sequence to generate a diagnosis set correlation sequence;

[0036] Step 103: Based on the medical item set sequence, calculate the medical item set correlation between any two adjacent medical visits to generate a medical item set correlation sequence;

[0037] Step 104: If the diagnosis set correlation sequence and the medical item set correlation sequence do not meet the preset conditions, it is determined that there is an abnormality in the target patient's medical record.

[0038] The method for identifying abnormal medical records provided in this embodiment can be applied to medical insurance fund supervision scenarios, medical institution behavior supervision scenarios, etc. This embodiment does not limit the application scenarios. The method for identifying abnormal medical records provided in this embodiment can more accurately and efficiently identify abnormal medical records, thereby promptly discovering abnormal behaviors such as borrowing medical insurance cards and falsifying medical records. The target patient mentioned in this embodiment can be any patient who has a medical consultation. Each visit of the patient can generate a medical record. The medical record usually records the patient's disease manifestations and the current diagnosis and treatment situation and other information in accordance with the standards. That is, the medical record usually contains rich and comprehensive information related to the medical consultation.

[0039] Based on this, in step 100, various text recognition technologies can be used to extract the diagnosis set and medical item set contained in the medical records of the target patient within a preset time. For example, all medical records within one year in the medical insurance system can be organized according to the patient ID, so as to determine the medical records corresponding to each patient and extract the diagnosis set and medical item set from them. Among them, the diagnosis set can be used to record the diagnostic results that occurred during the medical treatment process; the medical item set can be used to record the medical means consumed during the medical treatment process, which can usually be reflected in the charge details. A medical record may contain multiple diagnosis sets, such as a primary diagnosis set, a secondary diagnosis set, etc.; it may also contain multiple medical item sets, such as the used drug items, surgical operation items, nursing items, etc.

[0040] refer to Figure 1 and Figure 2a In step 100, the diagnosis sets and medical item sets occurring in each visit can be sorted according to the visit time to obtain a diagnosis set sequence and a medical item set sequence. The elements in the diagnosis set sequence are all the diagnosis sets occurring in a single visit, and the elements in the medical item set sequence are all the medical item sets occurring in a single visit.

[0041] In this embodiment, in the process of constructing a diagnosis set sequence and a medical item set sequence for a target patient, the diagnosis set and medical item set that occurred in each visit can be determined; the diagnosis set is sorted according to the visit time to obtain a diagnosis set sequence; the medical item set is sorted according to the visit time to obtain a medical item set sequence. Among them, within the diagnosis set and the medical item set, the elements (i.e., the diagnosis set or the medical item set) can be sorted according to the time of occurrence. In the case of the same occurrence time, they can be sorted according to the order of the records, or they can be sorted randomly. In addition, considering that a medical record may contain one or more diagnosis sets, the number of elements in the diagnosis set that occurs in a single visit can be one or more, and similarly, the number of elements in the medical item set can also be one or more.

[0042] For example, patient A has three visits to the doctor. The diagnosis set in the first visit record includes: a11, a12, a13; the diagnosis set in the second visit record includes: a21, a22; the diagnosis set in the third visit record includes: a31. Then the diagnosis set sequence of patient A will be {

a11,a12,a13

a21,a22

a31

a11,a12,a13

[0043] On this basis, in steps 102 and 103, the diagnosis set correlation between any two adjacent medical visits can be calculated based on the diagnosis set sequence to generate a diagnosis set correlation sequence; the medical item set correlation between any two adjacent medical visits can be calculated based on the medical item set sequence to generate a medical item set correlation sequence. Two adjacent medical visits refer to two medical visits that are adjacent in time. Continuing with the above example, the diagnosis set sequence for patient A contains diagnosis sets corresponding to three medical visits, where [a11, a12, a13] and [a21, a22] are adjacent, and [a21, a22] and [a31] are adjacent. The correlation between [a11, a12, a13] and [a21, a22], as well as the correlation between [a21, a22] and [a31], can be calculated. Thus, two diagnosis set correlation values ​​will be obtained for the diagnosis set sequence for patient A, and these two diagnosis set correlation values ​​can constitute the diagnosis set correlation sequence corresponding to patient A. In the diagnosis set correlation sequence, the elements represent the diagnosis set correlations between adjacent visits. Furthermore, the elements can be sorted according to the order of the related diagnosis sets in the diagnosis set sequence. Of course, this is not required, and this embodiment does not impose any restrictions on the order of the elements in the diagnosis set correlation sequence. Similarly, a medical item set correlation sequence for patient A can also be generated.

[0044] Among them, the technical solutions for the correlation of diagnosis sets and medical item sets will be described in detail later.

[0045] After generating the diagnosis set correlation sequence and medical item set correlation sequence corresponding to the target patient, refer to Figure 1 and Figure 2a In step 104, if the diagnosis set correlation sequence and the medical item set correlation sequence do not meet the preset conditions, it is determined that the target patient's medical record has an abnormality. The preset condition can be that the correlation representative value corresponding to the diagnosis set correlation sequence and the medical item set correlation sequence is not lower than the preset correlation threshold and the corresponding volatility is consistent with the specified volatility range. Accordingly, in step 104, the correlation representative value corresponding to the diagnosis set correlation sequence and the medical item set correlation sequence can be calculated respectively, wherein the correlation representative value adopts the correlation mean or median; the volatility corresponding to the diagnosis set correlation sequence and the medical item set correlation sequence can also be calculated; if the correlation representative value corresponding to the diagnosis set correlation sequence and the medical item set correlation sequence is not lower than the preset correlation threshold and the corresponding volatility is consistent with the specified volatility range, it can be determined that the target patient has no abnormal medical record; otherwise, it can be determined that the target patient's medical record has an abnormality.

[0046] Figure 3 This is a schematic diagram of the analysis results corresponding to the correlation sequences of several patients provided by an exemplary embodiment of this application. Figure 3 , the sequence volatility in Figure A is within the specified range and is relatively stable; and the values ​​of each element (correlation) in the sequence are relatively high. Therefore, it can be determined that the patient corresponding to Figure A has no abnormal medical records. The sequence volatility in Figure B is within the specified range and is relatively stable; however, the values ​​of each element (correlation) in the sequence are relatively low and do not meet the threshold requirements. Therefore, it can be determined that the medical records of the patient corresponding to Figure B are abnormal. The sequence volatility in Figure C exceeds the specified range and is very unstable; and the values ​​of each element (correlation) in the sequence are high and low, which does not meet the threshold requirements. Therefore, it can be determined that the medical records of the patient corresponding to Figure C are abnormal.

[0047] In this embodiment, a technical abstraction is employed to address the problem of identifying abnormal medical records. Each patient's medical records are sorted by visit time, generating a sequence of diagnosis sets and medical item sets. The correlation between two consecutive visits is then calculated, yielding a correlation sequence. For a normal patient, two consecutive visits are generally correlated when the time interval is short and the illnesses are not sudden (e.g., trauma), epidemic (e.g., influenza), or common (e.g., colds). For example, if the diagnosis at the previous visit was hypertension, then the current visit is likely related to a hypertension-related illness. In this embodiment, this correlation is also reflected in the diagnosis sets and medical item sets consumed during each visit. In other words, the diagnosis sets and medical item sets between the two visits also exhibit a high correlation, and the correlation sequence should be relatively stable with a high representative correlation value. Of course, occasional uncorrelated situations are normal. However, when medical records are intentionally forged, the diagnosis sets and medical item sets in the forged records often vary widely, making it difficult to maintain correlation between the forged records and other normal records. To this end, in this embodiment, preset conditions are configured for the correlation sequence. By judging whether the correlation sequence meets the preset conditions, it can be determined whether there is any abnormality in the patient's medical record.

[0048] In summary, in this embodiment, the patient is used as the identification unit, and based on the diagnosis set and medical item set contained in the patient's medical records within a preset time, the diagnosis set sequence and medical item set sequence corresponding to the patient are constructed according to the medical time; the diagnosis set correlation and medical item set correlation between two adjacent medical visits are calculated in sequence within the diagnosis set sequence and the medical item set sequence respectively to generate a diagnosis set correlation sequence and a medical item set correlation sequence; these two correlation sequences can be used as the basis for identifying abnormal medical records. If these two correlation sequences do not meet the preset conditions, it can be determined that the current patient's medical records are abnormal. In this way, based on the aforementioned diagnosis set correlation sequence and medical item set correlation sequence, it is possible to efficiently / accurately perceive whether there is an abnormal correlation between multiple medical visits of the patient, thereby efficiently / accurately realizing the identification of abnormal medical records.

[0049] In the above or following embodiments, adjacent medical item sets in a medical item set sequence can be traversed in order; the correlation between the currently traversed first medical item set and the second medical item set is calculated, where the first medical item set and the second medical item set are respectively the medical item sets corresponding to any two adjacent medical visits of the target patient; and other adjacent medical item sets in the medical item set sequence are continued to be traversed to sequentially generate the correlation between each adjacent medical item set in the medical item set sequence to form a medical item set correlation sequence.

[0050] Figure 4 A schematic diagram of a calculation scheme for the relevance of a medical project set provided by an exemplary embodiment of the present application. Figure 4 The circular pattern in the figure represents a single medical item set in the medical item set sequence. The correlation calculation operation can be performed on adjacent medical item sets in the medical item set sequence of the target patient in turn, thereby generating a medical item set correlation sequence.

[0051] In this embodiment, in-order traversal refers to traversing adjacent medical item sets in the time order of the medical item sets in the medical item set sequence. For example, if the medical item set sequence of patient B is {[b11, b12], [b21, b22, b23], [b31, b32, b33], [b41, b42]}, then the medical item set correlation 1 between [b11, b12] and [b21, b22, b23], the medical item set correlation 2 between [b21, b22, b23] and [b31, b32, b33], and the medical item set correlation 3 between [b31, b32, b33] and [b41, b42] can be calculated in sequence. In this way, the three medical item set correlations corresponding to patient B can be obtained in sequence and form the medical item set correlation sequence [medical item set correlation 1, medical item set correlation 2, medical item set correlation 3].

[0052] In this embodiment, the correlation between medical item sets can be measured based on the string similarity between the names of the medical item sets. However, due to the problem of name diversity in medical item sets, the name similarity between related medical item sets may not be high. For example, the two diagnostic names "renal failure" and "hemodialysis" are completely dissimilar, but in fact, these two diagnoses are very related. For example, the name strings of "irbesartan hydrochlorothiazide tablets" and "glimepiride tablets" are two drugs for treating hypertension. They are basically dissimilar, but both are commonly used drugs for treating hypertension. Therefore, measuring the correlation between medical item sets based on the name similarity between medical item sets may have the problem of insufficient accuracy. For this reason, it is proposed in this embodiment that the medical item sets can be vectorized, so that the correlation between medical item sets can be evaluated from the practical significance of the medical item sets.

[0053] Figure 2b This is a logical view of an optional implementation method provided by an exemplary embodiment of the present application. Figure 2b ,In this embodiment, entity vectors corresponding to various medical ,items can be pre-built based on global data.

[0054] In an exemplary scheme, the process of constructing the entity vector of a medical item may be: obtaining the medical records of several patients as samples, with each patient corresponding to at least one medical record; constructing a sample sequence of medical items consumed in each sample based on the medical records; inputting the sample sequence of medical items consumed in each sample into a vector construction model so that the vector construction model can capture the coexistence and / or co-accompaniment relationship between the medical items; and mapping each medical item to an adapted entity vector based on the captured coexistence and / or co-accompaniment relationship.

[0055] In this exemplary solution, the medical records of several patients can be used as global data. For example, all medical records from the past year can be used as samples. For the medical item collection, the medical items consumed during each visit are collected and a corresponding medical item sample sequence is constructed. This generates a medical item sample sequence for each visit. The medical items in each sample sequence can be sorted by occurrence time. For cases with the same occurrence time, they can be sorted by the order of the records, or randomly sorted. Based on this, the several medical item sample sequences generated based on the sample can be input into a vector construction model. In this embodiment, the vector construction model can utilize a language processing model such as a word2vec model, a bidirectional encoding BERT model, or a Transformer model. Because this exemplary solution constructs a special input sequence—that is, a medical item sample sequence constructed based on a medical visit—the vector construction model can capture co-occurrence and / or co-occurrence relationships between medical items from these special input sequences. A co-occurrence relationship can refer to medical items appearing in the same sequence, while a co-occurrence relationship can refer to medical items that do not co-occur but appear with one or more medical items.

[0056] For example:

[0057] 1) If XAXBX appears many times (X is a variable sequence of 0 to K elements), then the entity vectors of medical items A and B will have high similarity, meaning they co-occur in the same sequence. For example, if "irbesartan hydrochlorothiazide tablets" and "glimepiride tablets" frequently appear in the same sequence, then "irbesartan hydrochlorothiazide tablets" and "glimepiride tablets" are very related medical items, as they are both medications for treating hypertension.

[0058] 2) If XACX appears many times, and XBCX also appears many times (X is a variable sequence of 0 to K elements), then medical items A and B often share the same context (the latter is the context, the former is the context), and the entity vector similarity between medical items A and B is high. In other words, although medical items A and B do not co-occur, they both appear with medical item C, thus establishing a co-occurrence relationship between medical items A and B.

[0059] In this way, entity vectors can be used to represent the actual semantics of medical items, thereby more accurately and reasonably representing the actual meaning of medical items. In the process of constructing entity vectors, not only the actual meaning of medical items is understood from the perspective of a single patient and a single visit, but also the global data of all patients is integrated to more accurately understand the true meaning of medical items. This can more accurately represent medical items, thereby providing a more accurate basis for the subsequent calculation of the relevance of medical item sets.

[0060] In addition, the similarity (or distance) between entity vectors is related to the strength of the co-occurrence and / or co-accompaniment relationship between medical items. In this example, medical items with higher co-occurrence and / or co-accompaniment strength have more related entity vectors.

[0061] refer to Figure 2b Based on this, in this embodiment, when calculating the correlation between the currently traversed first and second medical item sets, the entity vectors corresponding to each medical item in the first and second medical item sets can be obtained; based on the entity vectors of the medical items, the correlation between the first and second medical item sets is calculated. In this way, during the process of identifying abnormal medical treatment, it is no longer necessary to calculate the entity vectors of the medical items in real time. Instead, the entity vectors pre-mapped for each medical item can be searched to obtain the entity vectors corresponding to each medical item in the first and second medical item sets.

[0062] In this embodiment, various implementations may be used to calculate the correlation between the currently traversed first medical item set and the second medical item set based on the entity vectors of the medical items.

[0063] In one implementation, the correlation between the entity vector corresponding to each medical item in the first medical item set and the entity vector corresponding to each medical item in the second medical item set can be calculated respectively; for each medical item in the first medical item set, the most relevant medical item matched in the second medical item set and the corresponding highest correlation are searched; for each medical item in the second medical item set, the most relevant medical item matched in the first medical item set and the corresponding highest correlation are searched; based on the highest correlation found for each medical item in the first medical item set, a first correlation of the first medical item set to the second medical item set is calculated; based on the highest correlation found for each medical item in the second medical item set, a second correlation of the second medical item set to the first medical item set is calculated; and based on the first correlation and the second correlation, the correlation between the first medical item set and the second medical item set is determined.

[0064] In calculating the first correlation between the first medical item set and the second medical item set based on the highest correlation found for each medical item in the first medical item set, the sum of the highest correlations found for each medical item in the first medical item set can be calculated; the ratio of the sum to the number of elements in the first medical item set can be used as the first correlation; and the second correlation can be calculated in the same manner. Furthermore, in determining the correlation between the first medical item set and the second medical item set based on the first and second correlations, the mean or maximum value between the first and second correlations can be used as the correlation between the first and second medical item sets.

[0065] For example, for the medical project set SetA and the medical project set SetB, to evaluate the correlation between SetB and SetA, then based on the entity vector, for each element b in SetB, find the most relevant element a from SetA, so that each element in SetB finds the most relevant element from SetA, and then add these correlations and divide them by the size of SetB to obtain the correlation SimBA between SetB and SetA. Similarly, for each element a in SetA, find the most relevant element b from SetB, so that each element in SetA finds the most relevant element from SetB, and then add these correlations and divide them by the size of SetA to obtain the correlation SimAB between SetA and SetB. Finally, the correlation between the two medical project sets is (SimAB+SimBA) / 2 or the larger value between the two. It should be understood that if the most relevant element of element b1 in SetB in SetA is a1, the most relevant element of element a1 in SetA in SetB is not necessarily b1. Therefore, in this implementation method, the correlation between medical project sets from different directions is integrated, and the correlation between medical project sets can be measured more accurately.

[0066] In another implementation: the first medical item set and the second medical item set can be regarded as two subsets in a bipartite graph, and the medical items included in the first medical item set and the second medical item set can be regarded as vertices in the bipartite graph; the vertex connectivity relationship between the two subsets in the bipartite graph is determined to obtain multiple connected branches contained in the bipartite graph; the vertex correlation is calculated within the multiple connected branches respectively; the correlation between the first medical item set and the second medical item set is determined based on the vertex correlation calculated within the multiple connected branches; wherein, if, based on the entity vector, it is determined that the correlation between the first medical item in the first medical item set and its most relevant second medical item in the second medical item set is not lower than a specified threshold, then it is determined that the first medical item and the second medical item are connected.

[0067] In the process of determining the vertex connectivity relationship between two subsets in a bipartite graph, the correlation between the entity vectors corresponding to each medical item in the first medical item set and the entity vectors corresponding to each medical item in the second medical item set can be calculated respectively; for each medical item in the first medical item set, the most relevant medical item assigned to it in the second medical item set can be searched; if the corresponding highest correlation is not lower than the specified threshold, the connectivity relationship between the corresponding vertices can be established; similarly, for each medical item in the second medical item set, the most relevant medical item assigned to it in the first medical item set can be searched; if the corresponding highest correlation is not lower than the specified threshold, the connectivity relationship between the corresponding vertices can be established.

[0068] Figure 5 This is a schematic diagram of an implementation method for calculating the correlation between medical project sets provided by an exemplary embodiment of the present application. Figure 5 In this implementation, medical item sets SetA and SetB are treated as two subsets in a bipartite graph, and the elements (medical item sets) in SetA and SetB are treated as vertices in the bipartite graph. Based on this, we can:

[0069] 1) For each element a in SetA, find the most relevant element b in SetB, and if the correlation is not less than the threshold sim_threshod, then construct a line between a and b;

[0070] 2) For each element b in SetB, find the most relevant element a in SetA, and if the correlation is not less than the threshold sim_threshod, then construct a line between a and b;

[0071] 3) Obtain all connected branches of the bipartite graph through 1) and 2), and the elements in the connected branches are directly or indirectly connected. Figure 5 Finally, 5 connected branches are obtained.

[0072] Alternatively, an exemplary approach to calculating vertex correlations within multiple connected branches may be to use the maximum correlation, minimum correlation, or average correlation between vertices within the connected branches as the vertex correlation within the connected branches. Furthermore, an exemplary approach to determining the correlation between the first and second medical item sets based on the vertex correlations calculated within the multiple connected branches may be to calculate the sum of the vertex correlations corresponding to the multiple connected branches, and to use the ratio of the sum to the number of connected branches as the correlation between the first and second medical item sets.

[0073] For example, continue to refer to Figure 5, the correlation can be calculated within each connected branch, and the correlation between the medical item set SetA and the medical item set SetB can be obtained by adding the correlations of multiple connected branches and dividing by the number of connected branches. Among them, the calculation method of calculating the correlation within the connected branch can be expressed by the maximum / minimum / average value of the correlation. Assume that the maximum value is used to represent the internal of the connected branch. Figure 5 There are 5 connected branches, and sim_threshod is 0.5. The correlation between SetA and SetB is: (max{0.85}+max{0.9,0.86}+max{0.93,0.95,0.88}) / 5=(0.85+0.9+0.95) / 5=0.54.

[0074] It is worth noting that the above two implementation methods of calculating the correlation between the currently traversed first medical item set and the second medical item set are exemplary, and this embodiment is not limited to this. Other implementation methods can also be used to calculate the correlation between the currently traversed first medical item set and the second medical item set, which will not be enumerated here.

[0075] In this way, by sequentially traversing the medical item set sequence of the target patient, a medical item set correlation sequence that can reflect the correlation of the medical item sets between each visit can be obtained.

[0076] In addition, as mentioned in the previous embodiment, in addition to constructing a medical item set sequence for a target patient, a diagnosis set sequence can be constructed for the target patient, and a diagnosis set relevance sequence can be generated based on the diagnosis set sequence. In this embodiment, the same technical solution as that implemented for the medical item set can be used to generate the diagnosis set relevance sequence corresponding to the target patient. However, the implementation scheme of constructing entity vectors for diagnoses in advance may be slightly different from the aforementioned implementation scheme of constructing entity vectors for medical items in advance. The main difference is that in the process of constructing entity vectors for diagnoses, the sample sequence is constructed on a patient basis instead of a visit basis, that is: the medical records of several patients can be obtained as samples, and each patient corresponds to at least one medical record; the diagnosis set appearing in at least one medical record of the patient is sorted according to the visit time, and the multiple diagnoses in each medical record are sorted according to the order in which the doctor wrote the diagnosis, or according to the primary and secondary order (primary diagnosis secondary diagnosis), or the time when the doctor made the diagnosis, to generate a diagnosis sample sequence corresponding to the patient; the diagnosis sample sequences corresponding to several patients are input into the vector construction model so that the vector construction model can capture the coexistence and / or co-accompaniment relationship between diagnoses; based on the captured coexistence and / or co-accompaniment relationship, each diagnosis is mapped to an adapted entity vector. For example, consider patient A's three medical records as samples. The first visit includes the diagnosis set timeA1: a11, a12, a13; the second visit includes the diagnosis set timeA2: a21, a22; and the third visit includes the diagnosis set timeA3: a31. Therefore, the diagnostic sample sequence for patient A is [a11, a12, a13, a21, a22, a31]. For patient B, consider two medical records as samples: the first visit includes the diagnosis set timeB1: b11; the second visit includes the diagnosis set timeB2: b21, a22. Therefore, the diagnostic sample sequence for patient B is [b11, b21, b22]. This allows for the construction of a diagnostic sample sequence for each patient, resulting in n diagnostic sample sequences for n patients. Based on these constructed special sequences, the diagnostic sample sequences corresponding to all patients can be used as input to the vector construction model. The entity vector can then output the entity vector for each diagnosis according to the internal logic mentioned above. Here, the patient is used as the unit to create the sample sequence for the vector construction model. This is mainly because the number of diagnoses in a single visit is generally small, and most cases only have one diagnosis. If the visit is still used as the processing unit, most diagnostic sample sequences will contain only one element, resulting in poor entity vector construction results.

[0077] To save space, the following only briefly describes the process of generating a diagnosis set relevance sequence based on a diagnosis set sequence. For relevant technical details, please refer to the description of the technical solution adopted for the medical project set above, but this should not cause any loss of the protection scope of this application.

[0078] In this embodiment, adjacent diagnosis sets in the diagnosis set sequence can be traversed in order; the correlation between the currently traversed first diagnosis set and the second diagnosis set is calculated, where the first diagnosis set and the second diagnosis set are the diagnosis sets corresponding to any two adjacent visits of the target patient; and other adjacent diagnosis sets in the diagnosis set sequence are continued to be traversed to generate the correlation between each adjacent diagnosis set in the diagnosis set sequence in order to form a diagnosis set correlation sequence.

[0079] In the process of calculating the correlation between the currently traversed first diagnosis set and the second diagnosis set: the entity vector corresponding to each diagnosis in the first diagnosis set and the second diagnosis set can be obtained, the entity vector is used to represent the actual semantics of the diagnosis, and diagnoses with higher co-occurrence and / or co-accompanying relationship strength have more relevant entity vectors; based on the entity vector of the diagnosis, the correlation between the first diagnosis set and the second diagnosis set is calculated.

[0080] In one implementation of calculating the correlation between a first diagnostic set and a second diagnostic set: the correlation between the entity vectors corresponding to each diagnosis in the first diagnostic set and the entity vectors corresponding to each diagnosis in the second diagnostic set can be calculated respectively; for each diagnosis in the first diagnostic set, the most relevant diagnosis matched in the second diagnostic set and the corresponding highest correlation are searched; for each diagnosis in the second diagnostic set, the most relevant diagnosis matched in the first diagnostic set and the corresponding highest correlation are searched; based on the highest correlation found for each diagnosis in the first diagnostic set, a first correlation of the first diagnostic set to the second diagnostic set is calculated; based on the highest correlation found for each diagnosis in the second diagnostic set, a second correlation of the second diagnostic set to the first diagnostic set is calculated; based on the first correlation and the second correlation, the correlation between the first diagnostic set and the second diagnostic set is determined.

[0081] Among them, in the process of calculating the first correlation of the first diagnostic set to the second diagnostic set based on the highest correlation found for each diagnosis in the first diagnostic set: the sum of the highest correlations found for each diagnosis in the first diagnostic set can be calculated; the ratio between the sum and the number of elements in the first diagnostic set is used as the first correlation, and the second correlation can be calculated in the same way; in the process of determining the correlation between the first diagnostic set and the second diagnostic set based on the first correlation and the second correlation: the mean or maximum value between the first correlation and the second correlation can be used as the correlation between the first diagnostic set and the second diagnostic set.

[0082] In another implementation method of calculating the correlation between the first diagnostic set and the second diagnostic set: the first diagnostic set and the second diagnostic set can be used as two subsets in a bipartite graph, and the diagnostic sets contained in the first diagnostic set and the second diagnostic set can be used as vertices in the bipartite graph; the vertex connectivity relationship between the two subsets in the bipartite graph is determined to obtain multiple connected branches contained in the bipartite graph; the vertex correlation is calculated within the multiple connected branches respectively; the correlation between the first diagnostic set and the second diagnostic set is determined based on the vertex correlation calculated within the multiple connected branches; wherein, if, based on the entity vector, the correlation between the first diagnosis in the first diagnostic set and its most relevant second diagnosis in the second diagnostic set is determined to be not less than a specified threshold, then it is determined that the first diagnosis and the second diagnosis are connected.

[0083] In which, in the process of calculating the vertex correlation within multiple connected branches respectively: the maximum correlation or the average correlation between the vertices within the connected branches can be used as the vertex correlation within the connected branches; in the process of determining the correlation between the first diagnostic set and the second diagnostic set based on the vertex correlation calculated within the multiple connected branches: it can be calculated as the sum of the vertex correlations corresponding to the multiple connected branches; the ratio between the sum and the number of connected branches is used as the correlation between the first diagnostic set and the second diagnostic set.

[0084] In summary, in this embodiment, the medical item set sequence and the diagnosis set sequence can be traversed as needed, and by vectorizing the diagnoses and medical items, the entity vectors can be used as the basis for calculating the correlation. In this way, the diagnosis set correlation and the medical item set correlation between each visit can be measured more accurately, thereby obtaining a more reasonable and accurate diagnosis set correlation sequence and medical item set correlation sequence, thereby improving the recognition efficiency and accuracy of abnormal medical records.

[0085] It should be noted that in some of the processes described in the above embodiments and the accompanying drawings, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different relevance, sets, etc., and do not represent a sequence of precedence, nor do they limit "first" and "second" to being different types.

[0086] Figure 6 This is a schematic diagram of a computing device provided by another exemplary embodiment of the present application. Figure 6As shown, the computing device includes a memory 60 and a processor 61 .

[0087] The processor 61 is coupled to the memory 60 and is configured to execute the computer program in the memory 60 to:

[0088] Extract diagnosis sets and medical item sets from the target patient's medical records within a preset time period, with each medical visit corresponding to one medical record.

[0089] According to the time of the consultation, the diagnosis set and medical item set occurring in each consultation are sorted respectively to obtain the diagnosis set sequence and the medical item set sequence;

[0090] Based on the diagnosis set sequence, the diagnosis set correlation between any two adjacent visits is calculated to generate a diagnosis set correlation sequence;

[0091] Based on the medical item set sequence, calculating the medical item set correlation between any two adjacent medical visits; to generate a medical item set correlation sequence;

[0092] If the diagnosis set correlation sequence and the medical item set correlation sequence do not meet the preset conditions, it is determined that there is an abnormality in the target patient's medical record.

[0093] In an optional embodiment, the processor 61 calculates the medical item set correlation between any two adjacent medical visits based on the medical item set sequence to generate the medical item set correlation sequence, which may be used to:

[0094] Traverse the adjacent medical item sets in the medical item set sequence in order;

[0095] Calculate the correlation between the first medical item set and the second medical item set currently traversed, where the first medical item set and the second medical item set are medical item sets corresponding to any two adjacent medical visits of the target patient;

[0096] Continue to traverse other adjacent medical item sets in the medical item set sequence to sequentially generate correlations between adjacent medical item sets in the medical item set sequence to form a medical item set correlation sequence.

[0097] In an optional embodiment, during the process of calculating the correlation between the currently traversed first medical item set and the second medical item set, the processor 61 may be configured to:

[0098] Obtaining entity vectors corresponding to respective medical items in the first medical item set and the second medical item set, wherein the entity vectors corresponding to the medical items are used to represent the actual semantics of the medical items, and medical items with higher co-occurrence and / or co-accompaniment relationship strengths have more relevant entity vectors;

[0099] Based on the entity vectors of the medical items, a correlation between the first medical item set and the second medical item set is calculated.

[0100] In an optional embodiment, when calculating the relevance between the first medical item set and the second medical item set based on the entity vectors of the medical items, the processor 61 may be configured to:

[0101] respectively calculating the correlation between the entity vectors corresponding to each medical item in the first medical item set and the entity vectors corresponding to each medical item in the second medical item set;

[0102] Searching for each medical item in the first medical item set the most relevant medical item matched in the second medical item set and the corresponding highest correlation;

[0103] searching for the most relevant medical item and the corresponding highest correlation degree for each medical item in the second medical item set in the first medical item set;

[0104] calculating a first correlation between the first medical item set and the second medical item set based on the highest correlation found for each medical item in the first medical item set;

[0105] calculating a second correlation of the second medical item set to the first medical item set based on the highest correlation found for each medical item in the second medical item set;

[0106] The correlation between the first medical item set and the second medical item set is determined according to the first correlation and the second correlation.

[0107] In an optional embodiment, when calculating the first correlation between the first medical item set and the second medical item set based on the highest correlation found for each medical item in the first medical item set, the processor 61 may be configured to:

[0108] Calculated as the sum of the highest correlations found for each medical item in the first medical item set;

[0109] The ratio between the sum value and the specification of the first medical project set is used as the first correlation;

[0110] In the process of determining the correlation between the first medical item set and the second medical item set according to the first correlation and the second correlation, it can be used to:

[0111] The average or maximum value between the first correlation and the second correlation is used as the correlation between the first medical item set and the second medical item set.

[0112] In an optional embodiment, when calculating the relevance between the first medical item set and the second medical item set based on the entity vectors of the medical items, the processor 61 may be configured to:

[0113] The first medical item set and the second medical item set are used as two subsets in a bipartite graph, and the medical items included in the first medical item set and the second medical item set are used as vertices in the bipartite graph;

[0114] Determine the vertex connectivity relationship between two subsets in the bipartite graph to obtain multiple connected branches contained in the bipartite graph;

[0115] Calculate vertex correlation within multiple connected branches respectively;

[0116] determining a correlation between the first medical item set and the second medical item set according to vertex correlations calculated within the plurality of connected branches;

[0117] If, based on the entity vector, it is determined that the correlation between a first medical item in the first medical item set and its most relevant second medical item in the second medical item set is not less than a specified threshold, then the first medical item and the second medical item are determined to be connected.

[0118] In an optional embodiment, when the processor 61 calculates vertex correlations within a plurality of connected branches, the processor 61 may be configured to: use the maximum correlation or the average correlation between vertices within the connected branches as the vertex correlation within the connected branches;

[0119] In the process of determining the correlation between the first medical item set and the second medical item set based on the vertex correlations calculated within the multiple connected branches, it can be used to:

[0120] Calculated as the sum of the correlations between vertices corresponding to multiple connected branches;

[0121] The ratio of the sum value to the number of connected branches is used as the correlation between the first medical item set and the second medical item set.

[0122] In an optional embodiment, during the process of constructing the entity vector of the medical item, the processor 61 may be used to:

[0123] Obtain medical records of several patients as samples, with each patient corresponding to at least one medical record;

[0124] Taking medical visits as the unit, construct the sample sequence of medical items consumed in each sample;

[0125] Inputting the sample sequences of medical items consumed in each sample into a vector construction model, so that the vector construction model can capture the co-existence and / or co-accompaniment relationship between the medical items;

[0126] Based on the captured co-occurrence and / or co-accompaniment relationships, each medical item is mapped to an adapted entity vector.

[0127] In an optional embodiment, during the process of obtaining the entity vector corresponding to each medical item in the first medical item set and the second medical item set, the processor 61 may be configured to:

[0128] The entity vector pre-mapped for each medical item is searched to obtain the entity vector corresponding to each medical item in the first medical item set and the second medical item set.

[0129] In an optional embodiment, the vector construction model adopts a word2vec model, a bidirectional encoding BERT model, or a transform Transformer model.

[0130] In an optional embodiment, when calculating the diagnosis set correlation between any two adjacent visits based on the diagnosis set sequence to generate the diagnosis set correlation sequence, the processor 61 may be configured to:

[0131] Traverse the adjacent diagnosis sets in the diagnosis set sequence in order;

[0132] Based on the entity vectors corresponding to each diagnosis in the currently traversed first diagnosis set and second diagnosis set, the correlation between the first diagnosis set and the second diagnosis set is calculated. The first diagnosis set and the second diagnosis set are the diagnosis sets corresponding to any two adjacent visits of the target patient. The entity vectors corresponding to the diagnoses are used to represent the actual semantics of the diagnoses, and diagnoses with higher co-occurrence and / or co-occurrence relationships have more correlated entity vectors.

[0133] Continue to traverse other adjacent diagnosis sets in the diagnosis set sequence to sequentially generate correlations between adjacent diagnosis sets in the diagnosis set sequence to form a diagnosis set correlation sequence.

[0134] In an optional embodiment, during the construction of the diagnostic entity vector, the processor 61 may be used to:

[0135] Obtain medical records of several patients as samples, with each patient corresponding to at least one medical record;

[0136] Taking patients as units, sort the diagnosis sets appearing in at least one of the patient's medical records according to the time of medical consultation to generate a diagnosis sample sequence corresponding to the patient;

[0137] Inputting diagnostic sample sequences corresponding to a plurality of patients into a vector construction model, so that the vector construction model can capture the coexistence and / or co-occurrence relationship between diagnoses;

[0138] Based on the captured co-occurrence and / or co-occurrence relations, each diagnosis is mapped to an adapted entity vector.

[0139] In an optional embodiment, the preset conditions include that the correlation representative values ​​corresponding to the diagnosis set correlation sequence and the medical item set correlation sequence are not lower than a preset correlation threshold and the fluctuation rates corresponding to the respective ones are within a specified fluctuation range;

[0140] The representative value of correlation is the mean or median of correlation.

[0141] Further, if Figure 6 As shown, the computing device also includes: a communication component 62, a power supply component 63 and other components. Figure 6 Only some components are shown schematically, and it does not mean that the computing device only includes Figure 6 Components shown.

[0142] It is worth noting that the technical details in the above-mentioned embodiments of the computing device can be referred to the relevant description in the aforementioned method embodiment. In order to save space, they will not be repeated here, but this should not cause any loss of the scope of protection of this application.

[0143] Accordingly, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed, can implement the steps that can be executed by a computing device in the above method embodiment.

[0144] above Figure 6 The memory in the computing platform is used to store computer programs and can be configured to store various other data to support operations on the computing platform. Examples of such data include instructions for any application or method operating on the computing platform, contact data, phone book data, messages, pictures, videos, etc. The memory can be implemented by any type of volatile or non-volatile storage device or a combination of them, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0145] above Figure 6The communication component is configured to facilitate wired or wireless communication between the device where the communication component is located and other devices. The device where the communication component is located can access a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G / LTE, 6G and other mobile communication networks, or a combination thereof. In an exemplary embodiment, the communication component receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0146] above Figure 6 The power supply component in a device provides power to various components of the device in which the power supply component is located. The power supply component may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which the power supply component is located.

[0147] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0148] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0149] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1The function specified in one or more boxes.

[0150] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0151] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0152] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0153] Computer-readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0154] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0155] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included in the protection scope of the present application.

Claims

1. A method for identifying abnormal medical records, comprising: Extract diagnosis sets and medical item sets from the target patient's medical records within a preset time period, with each medical visit corresponding to one medical record. According to the time of the consultation, the diagnosis set and medical item set occurring in each consultation are sorted respectively to obtain the diagnosis set sequence and the medical item set sequence; Based on the diagnosis set sequence, calculating the diagnosis set correlation between any two adjacent visits to generate a diagnosis set correlation sequence; Based on the medical item set sequence, calculating the medical item set correlation between any two adjacent medical visits to generate a medical item set correlation sequence; If the diagnosis set correlation sequence and the medical item set correlation sequence do not meet the preset conditions, it is determined that there is an abnormality in the medical record of the target patient.

2. The method according to claim 1, wherein the correlation between the medical item sets between any two adjacent medical visits is calculated based on the medical item set sequence; To generate a correlation sequence of medical project sets, including: Traversing the adjacent medical item sets in the medical item set sequence in order; Calculating the correlation between the first medical item set and the second medical item set currently traversed, where the first medical item set and the second medical item set are respectively medical item sets corresponding to any two adjacent medical visits of the target patient; Continue to traverse other adjacent medical item sets in the medical item set sequence to sequentially generate correlations between adjacent medical item sets in the medical item set sequence to form the medical item set correlation sequence.

3. The method according to claim 2, wherein calculating the correlation between the first medical item set and the second medical item set currently traversed comprises: Obtaining entity vectors corresponding to respective medical items in the first medical item set and the second medical item set, where the entity vectors corresponding to the medical items are used to represent the actual semantics of the medical items, and medical items with higher co-occurrence and / or co-accompaniment relationship strengths have more relevant entity vectors; Based on the entity vectors of the medical items, the correlation between the first medical item set and the second medical item set is calculated.

4. The method according to claim 3, wherein the calculating the correlation between the first medical item set and the second medical item set based on the entity vectors of the medical items comprises: respectively calculating the correlation between the entity vector corresponding to each medical item in the first medical item set and the entity vector corresponding to each medical item in the second medical item set; searching for the most relevant medical item and the corresponding highest correlation degree matched by each medical item in the first medical item set in the second medical item set; searching for the most relevant medical item and the corresponding highest correlation degree matched by each medical item in the second medical item set in the first medical item set; calculating a first correlation between the first medical item set and the second medical item set according to the highest correlation found for each medical item in the first medical item set; calculating a second correlation of the second medical item set to the first medical item set according to the highest correlation found for each medical item in the second medical item set; The correlation between the first medical item set and the second medical item set is determined according to the first correlation and the second correlation.

5. The method according to claim 4, wherein the calculating, based on the highest correlation found for each medical item in the first medical item set, a first correlation between the first medical item set and the second medical item set comprises: Calculated as the sum of the highest correlations found for each medical item in the first medical item set; taking the ratio between the sum value and the specification of the first medical item set as the first correlation; The determining the correlation between the first medical item set and the second medical item set according to the first correlation and the second correlation includes: The average or maximum value between the first correlation and the second correlation is used as the correlation between the first medical item set and the second medical item set.

6. The method according to claim 3, wherein the calculating the correlation between the first medical item set and the second medical item set based on the entity vectors of the medical items comprises: The first medical item set and the second medical item set are used as two subsets in a bipartite graph, and the medical items included in the first medical item set and the second medical item set are used as vertices in the bipartite graph; Determining a vertex connectivity relationship between two subsets in the bipartite graph to obtain a plurality of connected branches contained in the bipartite graph; Calculating vertex correlations within the plurality of connected branches respectively; determining a correlation between the first medical item set and the second medical item set according to vertex correlations calculated within the plurality of connected branches; If, based on the entity vector, it is determined that the correlation between a first medical item in the first medical item set and its most relevant second medical item in the second medical item set is not less than a specified threshold, then it is determined that the first medical item and the second medical item are connected.

7. The method according to claim 6, wherein the step of respectively calculating vertex correlations within the plurality of connected branches comprises: The maximum correlation or the mean correlation between the vertices in the connected branch is used as the vertex correlation within the connected branch; The determining the correlation between the first medical item set and the second medical item set based on the vertex correlations calculated within the plurality of connected branches includes: Calculate the sum of the correlations between the vertices corresponding to the multiple connected branches; The ratio of the sum value to the number of connected branches is used as the correlation between the first medical item set and the second medical item set.

8. The method according to claim 3, wherein the process of constructing the entity vector of the medical item comprises: Obtain medical records of several patients as samples, with each patient corresponding to at least one medical record; Taking medical visits as the unit, construct the sample sequence of medical items consumed in each sample; Inputting a sequence of medical item samples consumed in each sample into a vector construction model, so that the vector construction model can capture the co-existence and / or co-accompaniment relationship between the medical items; Based on the captured co-occurrence and / or co-accompaniment relationships, each medical item is mapped to an adapted entity vector.

9. The method according to claim 8, wherein obtaining the entity vector corresponding to each medical item in the first medical item set and the second medical item set comprises: The entity vector pre-mapped for each medical item is searched to obtain the entity vector corresponding to each medical item in the first medical item set and the second medical item set.

10. The method according to claim 1, wherein the correlation between the diagnosis sets of any two adjacent visits is calculated based on the diagnosis set sequence; To generate a diagnostic set correlation sequence, including: Traversing adjacent diagnosis sets in the diagnosis set sequence in order; Calculate the correlation between the first diagnosis set and the second diagnosis set based on the entity vectors corresponding to each diagnosis in the currently traversed first diagnosis set and the second diagnosis set, where the first diagnosis set and the second diagnosis set are the diagnosis sets corresponding to any two adjacent visits of the target patient. The entity vectors corresponding to the diagnoses are used to represent the actual semantics of the diagnoses, and diagnoses with higher co-occurrence and / or co-occurrence relationships have more correlated entity vectors. Continue to traverse other adjacent diagnosis sets in the diagnosis set sequence to sequentially generate correlations between adjacent diagnosis sets in the diagnosis set sequence to form the diagnosis set correlation sequence.

11. The method according to claim 10, wherein the process of constructing the entity vector of the diagnosis comprises: Obtain medical records of several patients as samples, with each patient corresponding to at least one medical record; Taking patients as units, sort the diagnosis sets appearing in at least one of the patient's medical records according to the time of medical consultation to generate a diagnosis sample sequence corresponding to the patient; Inputting diagnostic sample sequences corresponding to a plurality of patients into a vector construction model, so that the vector construction model can capture the coexistence and / or co-occurrence relationship between diagnoses; Based on the captured co-occurrence and / or co-occurrence relations, each diagnosis is mapped to an adapted entity vector.

12. The method according to claim 1, wherein the preset conditions include that the correlation representative values ​​corresponding to the diagnosis set correlation sequence and the medical item set correlation sequence are not lower than a preset correlation threshold and the fluctuation rates corresponding to the respective ones are within a specified fluctuation range; The correlation representative value adopts the correlation mean or median.

13. A computing device comprising a memory and a processor; The memory is used to store one or more computer instructions; The processor is coupled to the memory and configured to execute the one or more computer instructions for: Extract diagnosis sets and medical item sets from the target patient's medical records within a preset time period, with each medical visit corresponding to one medical record. According to the time of the consultation, the diagnosis set and medical item set occurring in each consultation are sorted respectively to obtain the diagnosis set sequence and the medical item set sequence; Based on the diagnosis set sequence, calculating the diagnosis set correlation between any two adjacent visits to generate a diagnosis set correlation sequence; Based on the medical item set sequence, calculating the medical item set correlation between any two adjacent medical visits to generate a medical item set correlation sequence; If the diagnosis set correlation sequence and the medical item set correlation sequence do not meet the preset conditions, it is determined that there is an abnormality in the medical record of the target patient.

14. A computer-readable storage medium storing computer instructions, which, when executed by one or more processors, causes the one or more processors to execute the method for identifying abnormal medical records according to any one of claims 1 to 12.

Images