Abnormal medical behavior recognition method, device and interaction system

Abstract

An abnormal medical behavior identification method, device and interaction system are disclosed. The method comprises: dividing patient treatment records of a medical institution by time to obtain multiple treatment time slices for the same medical institution; extracting accompanying behaviors from the treatment time slices and screening accompanying relationships that meet predetermined conditions from the extracted accompanying behaviors; and constructing an accompanying network based on the accompanying relationships and finding multiple patients having an accompanying relationship with each other from the accompanying network as a suspected group with abnormal medical behavior. The present application can automatically locate individuals with accompanying behaviors from massive data quickly and identify suspected groups with abnormal medical behavior according to two accompanying relationships, thereby improving the identification accuracy and efficiency of groups with truly abnormal behavior.

Description

Technical Field

[0001] This disclosure relates to the medical field, and more particularly to a method, apparatus, and interactive system for identifying abnormal medical behavior. Background Technology

[0002] With the continuous development of medical technology and the improvement of people's living standards, life expectancy is also increasing. As the aging population deepens, the pressure on the medical insurance fund is also growing.

[0003] Therefore, a solution is needed that uses technological means to accurately identify abnormal medical behaviors. Summary of the Invention

[0004] One technical problem this disclosure aims to solve is to provide an abnormal medical behavior identification scheme that can automatically and quickly locate individuals with accompanying behaviors from massive amounts of data, and can identify suspected groups with abnormal medical behaviors based on pairwise accompanying relationships, thereby improving the accuracy and efficiency of abnormal medical behavior identification.

[0005] According to a first aspect of this disclosure, a method for identifying abnormal medical behavior is provided, comprising: dividing patient visit records of a medical institution into time segments to obtain multiple visit time slices for the same medical institution; extracting accompanying behaviors from the visit time slices, and filtering out accompanying relationships that meet predetermined conditions from the extracted accompanying behaviors; and constructing an accompanying network based on the accompanying relationships, and searching for multiple patients with accompanying relationships with each other in the accompanying network as a group suspected of having abnormal medical behavior.

[0006] Optionally, dividing patient visit records at a medical institution into time slices to obtain multiple visit time slices for the same medical institution includes: sorting patient visit records at a specific medical institution on a specific date in chronological order; and dividing the visit records into visit time slices using a time window of equal to a predetermined duration that slides along the time flow.

[0007] Optionally, extracting accompanying behaviors from the consultation time slice and filtering out accompanying relationships that meet predetermined conditions from the extracted accompanying behaviors includes: determining that all patients in the same consultation time slice have one accompanying behavior in each pair; and determining that the two patients have an accompanying relationship when the number of accompanying behaviors between the two patients reaches a predetermined threshold number.

[0008] Optionally, constructing a companion network based on the companion relationship and searching for multiple patients with companion relationships in the companion network as suspected groups with abnormal medical behavior includes: representing patients as nodes and companion relationships as edges to construct a companion network; and identifying maximal cliques with edges connecting each pair of nodes in the companion network as suspected groups with abnormal medical behavior.

[0009] Optionally, the method includes at least one of the following: obtaining patient visit records from a certain medical institution, and extracting accompanying behaviors only from the visit time slices of that medical institution to determine the accompanying relationship; or obtaining patient visit records from all medical institutions within a specific region, and extracting accompanying behaviors from the visit time slices of the all medical institutions to determine the accompanying relationship; or obtaining patient visit records from medical institutions in a specific field within a specific region, and extracting accompanying behaviors from the visit time slices of the medical institutions in the specific field to determine the accompanying relationship.

[0010] Optionally, individuals with normal medical practices can be removed from groups suspected of having abnormal medical practices, based on the type of disease and the amount of consumption.

[0011] According to a second aspect of the present invention, an abnormal medical behavior identification device is provided, comprising: a time stream processing module for dividing patient visit records of a medical institution into time segments to obtain multiple visit time slices for the same medical institution; a companion relationship extraction module for extracting companion behaviors from the visit time slices and filtering companion relationships that meet predetermined conditions from the extracted companion behaviors; and a network construction and search module for constructing a companion network based on the companion relationships and searching for multiple patients with companion relationships with each other in the companion network as a group suspected of having abnormal medical behavior.

[0012] Optionally, the device further includes an abnormal medical expense alarm module, which alarms the medical expense items of patients in the group when the disease type and reimbursement amount of the group suspected of having abnormal medical behavior are not removed from the preset whitelist.

[0013] According to a third aspect of the present invention, an abnormal medical behavior recognition interactive system is proposed, comprising: a setting interface module for setting the range of patient medical records to be selected from the database for processing and setting the duration of the medical visit time slice; and a display interface module for displaying an accompanying network composed of accompanying behaviors extracted from the medical visit time slice that meet predetermined conditions, and highlighting multiple patient nodes in the accompanying network that have an accompanying relationship with each other, wherein the display interface includes an operable module for displaying the medical consumption records of the patient node when selected.

[0014] According to a fourth aspect of the invention, a non-transitory machine-readable storage medium is provided, on which executable code is stored, which, when executed by a processor of an electronic device, causes the processor to perform the method as described in the first aspect.

[0015] Therefore, the abnormal medical behavior identification scheme of the present invention can filter out patients with accompanying relationships from massive data, and quickly and accurately identify patients with group behavior by converting the above relationships into a graph, and determine them as suspected abnormal medical behavior groups, so as to make a judgment on real abnormal medical expenditures. Attached Figure Description

[0016] The above and other objects, features and advantages of this disclosure will become more apparent from the more detailed description of exemplary embodiments thereof taken in conjunction with the accompanying drawings, wherein like reference numerals generally denote like parts.

[0017] Figure 1 A schematic flowchart of an abnormal medical behavior identification method according to an embodiment of the present invention is shown.

[0018] Figure 2 An example of using time windows to divide medical records is shown.

[0019] Figure 3 A schematic diagram of a computing device that can be used to implement the above-described abnormal medical behavior identification method according to an embodiment of the present invention is shown.

[0020] Figure 4 A schematic diagram of the components of an abnormal medical behavior recognition device according to an embodiment of the present invention is shown.

[0021] Figure 5 A schematic flowchart of an interactive system for recognizing abnormal medical behavior according to an embodiment of the present invention is shown.

[0022] Figure 6 A schematic diagram illustrating the specific implementation steps of an abnormal medical behavior identification method according to a preferred embodiment of the present invention is shown. Detailed Implementation

[0023] Preferred embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While preferred embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0024] As mentioned earlier, gangs composed of multiple members engaging in fraudulent medical practices, purchasing and reselling drugs and supplies that do not match actual medical needs, often exhibit characteristics such as secrecy, professionalism, division of labor, large sums of money involved, and profound harm. When a gang is illegally obtaining medical insurance drugs and supplies, its members often visit the same medical institution simultaneously to commit insurance fraud. These actions frequently occur repeatedly.

[0025] In existing technologies, abnormal data needs to be manually screened and identified. This process is time-consuming and labor-intensive, and can lead to the omission of a large number of insurance fraud cases. Therefore, this invention proposes a data-driven automatic identification method for abnormal medical behavior. This method can identify all groups suspected of exhibiting abnormal medical behavior by accurately locating the accompanying behaviors among members, thereby assisting in further screening and investigation.

[0026] Figure 1 A schematic flowchart illustrating an abnormal medical behavior identification method according to an embodiment of the present invention is shown. This method can automatically mine massive amounts of medical records and accurately identify suspected abnormal medical behavior groups with accompanying relationships.

[0027] In step S110, the patient visit records of the medical institution are divided into time segments to obtain multiple visit time slices for the same medical institution.

[0028] In step S130, accompanying behaviors are extracted from the consultation time slices, and accompanying relationships that meet predetermined conditions are filtered from the extracted accompanying behaviors. Here, the occurrence of two patients within the same consultation time slice can be determined as indicating that there is one accompanying behavior between the two patients. If the accompanying behavior occurs abnormally frequently, for example, if it meets a predetermined threshold, then it can be determined that there is an accompanying relationship between the two patients.

[0029] In step S130, a companion network is constructed based on the companion relationship, and multiple patients who have companion relationships with each other are identified from the companion network as a group suspected of having abnormal medical behavior.

[0030] Therefore, it is possible to quickly locate individuals with accompanying behaviors from massive amounts of data in an automated manner, and to identify suspected groups with abnormal medical behaviors based on pairwise companion relationships, thereby improving the accuracy and efficiency of abnormal medical intervention.

[0031] In order to determine, as mentioned above, the situation where two patients fall into the same medical institution for the same predetermined time period as a case of two patients having a shared activity, it is necessary to know which patients came to the same medical institution for a certain predetermined time period. Therefore, the identification method of this invention needs to obtain the patient visit records of the medical institution and divide the visit records of the medical institution into predetermined time periods.

[0032] In the field of medical insurance, each patient's visit to a medical institution or purchase of medicines and consumables at a designated medical insurance pharmacy creates a record of their visit / purchase. This record includes the medical institution / pharmacy, patient ID, visit ID, visit time, and details of the payment items (including examinations, surgeries, medicines, consumables, nursing care, etc.). In the context of this invention, medical institutions can include not only hospitals and clinics capable of diagnosis, treatment, and prescription, but also designated medical insurance pharmacies. Furthermore, the visit time can be the time the patient registers, the time the doctor issues the prescription, or the time the patient picks up the medication from the pharmacy. In a preferred embodiment, one type of time is uniformly selected as the visit time, for example, the time the patient picks up medication from the pharmacy.

[0033] When a criminal gang commits fraud, they often visit the same medical institution multiple times within the same time period to defraud insurance companies. Therefore, the data used to identify such gangs can include medical institution ID, patient ID, consultation ID, and consultation time. The medical institution ID is used to distinguish different medical institutions, the patient ID, along with the consultation ID, can be used to uniquely identify a patient's consultation, and the consultation time can be used to determine which predetermined time period the patient's treatment belongs to and which patients during the same time period can be considered as having a concurrent relationship with the patient.

[0034] Here, patient medical records can be retrieved from the databases of medical institutions or management agencies. Since patient medical information involves patient privacy, when these records are processed by a third party outside the medical institution, the patient information needs to be anonymized. For example, the patient ID used to identify the patient in the medical record could be the patient's medical insurance card number or other unique ID number. In the actual implementation of the identification method of this invention, the patient ID can be processed, for example, by performing a hash transformation. This ensures that the data processor does not know the patient's real ID, but the same patient ID can correspond to the same hash code, thus enabling the identification of the same patient's medical visits. If the identification determines that the patient has exhibited abnormal behavior, the relevant agency can then retrieve the patient's real ID by querying, for example, a hash transformation mapping table. This ensures the security of the privacy information of other patients seeking medical treatment normally.

[0035] After obtaining the patient's medical records, these records can be segmented, for example, according to predetermined durations, thus obtaining time slices for determining accompanying behaviors. Since gang members typically concentrate their insurance fraud activities within a single time period, the predetermined duration is usually shorter than the daily opening hours of the medical institution. For example, the predetermined duration can be set to one and a half hours, two hours, etc.

[0036] In one embodiment, medical records can be directly divided according to a predetermined duration. For example, if the predetermined duration is two hours, a day's medical records can be divided into time slots of 8:00-10:00, 10:00-12:00, and so on. Patients falling into each of these time slots are then paired to determine if they have a communal relationship. While this division simplifies the division of predetermined time slots, multiple patients from a group, even if they visit within the same time slot, will not visit at exactly the same time, thus missing a significant number of members with a group relationship. For example, two patients who visit at 9:59 and 10:01 clearly belong to the same time slot, but the above division would cause the communal behavior between these two patients to be missed. Therefore, as a remedy, different predetermined time slots can be divided based on different starting points. For example, in the case above where a day's medical records are divided into time slots like 8:00-10:00, 10:00-12:00, and so on, and patients falling into each of these time slots are paired to determine if they have accompanying behaviors, we can also start from 9:00 and divide the day's medical records into time slots like 9:00-11:00, 11:00-13:00, and so on (treating 8-9 as a separate time slot), thus performing a second division. In this second division, two patients who visited at 9:59 and 10:01 can be assigned to the same 9-11:00 time slot. Subsequently, accompanying behaviors between patients can be determined based on this second division, and duplicate accompanying behaviors can be deleted.

[0037] In a preferred embodiment, the present invention can also utilize time windows to perform more precise and less omission-prone division of predetermined time periods. To this end, dividing the medical records of a medical institution for a predetermined duration includes: sorting patient medical records of a specific date at a specific medical institution in chronological order; and using a time window equal to the predetermined duration and sliding according to the time flow to perform the division of the medical records.

[0038] In this invention, a medical institution can refer to an institution that provides medical services and may be the location where abnormal medical behavior occurs, such as a hospital, clinic, or pharmacy.

[0039] Since members of a group often visit the same medical institution for treatment or to purchase medications (including supplies) at the same time, the situation where two patients visit the same medical institution for the same predetermined duration can be considered as two patients engaging in a shared activity. The predetermined duration can be determined based on the typical time required for a group member's shared activity (or slightly longer than a typical event). In one example, the predetermined duration could be two hours. Therefore, if patients A and B arrive at Hospital A at 10:00 AM and 10:30 AM respectively on October 8th, it can be considered that patients A and B engaged in a shared activity.

[0040] When the accompanying behaviors of two patients meet predetermined conditions, it is determined that the two patients have a companion relationship. This is because one or two instances of companionship between two patients may be coincidental, but if the two patients exhibit multiple behaviors (e.g., no less than 5 times, where the predetermined condition could be no less than a predetermined threshold number of times), it indicates that the two patients have a companion relationship, i.e., constitute a suspected gang relationship. For example, if patients A and B only visited the clinic at the same time on October 8th, and visited on different dates at other times in October, then the accompanying behavior of the two patients on October 8th can be considered to have occurred by chance. However, if patients A and B visit Hospital A at the same time on five working days in October (for example, patients A and B visit Hospital A at the same time on October 8th between 10-11 am, October 13th between 8-9 am, October 15th between 8-9 am, October 18th between 3-5 pm, and October 22nd between 1-2 pm), then it can be considered that there is a high probability that patients A and B have a comorbid relationship.

[0041] By constructing a companion network based on the aforementioned companion relationships, multiple patients who share these companion relationships can be identified from a pool of companion relationships. These patients are then grouped together as a group suspected of exhibiting abnormal medical behavior. For example, in the above case, not only patients A and B, but also patients C and D, all visited Hospital A simultaneously on October 8th between 10-11 AM, October 13th between 8-9 AM, October 15th between 8-9 AM, October 18th between 3-5 PM, and October 22nd between 1-2 PM. According to steps S110 and S120 above, companion relationships exist between patients C and D, patients A and C, patients A and D, patients B and C, and patients B and D. Therefore, patients A, B, C, and D can be identified as belonging to the same group suspected of exhibiting abnormal medical behavior.

[0042] Therefore, this invention can identify suspected groups by determining the association relationship, thereby improving the detection of groups that actually engage in abnormal insurance fraud.

[0043] Figure 2 An example of using time windows to segment medical records is shown. As illustrated, the visit times of patients at a particular medical institution on a given day can be sorted chronologically. Each point in the figure represents a patient's medical record. Then, a time window with a width of, for example, 2 hours (i.e., equal to the width of the predetermined duration) can be used to segment the records for the predetermined duration. Since the distance the time window slides each time is typically less than the length of the time window itself, when determining that two patients fall into the same medical institution for the same predetermined duration as having a shared activity, it is necessary to remove the duplicated shared activity caused by overlapping time window segmentation.

[0044] In one embodiment, the time window can be slid in relatively fixed steps, the steps being smaller than the predetermined duration. For example, the time window can be slid in steps of 15 minutes.

[0045] In a preferred embodiment of the present invention, the time window can slide N patient medical records at a time, and if the patient medical records in the window are the same as those in the previous window or are a subset of those records, then another N patient medical records can be slide, where N is an integer not less than 1.

[0046] In a highly accurate implementation, N equals 1. Each time, the system slides through patients from the very beginning of a hospital's operations, one patient at a time. If a patient in the current window is identical to or a subset of the patients in the previous window, then another patient is slided through. This operation ensures no accompanying actions are missed, but windows may overlap. To prevent overlapping patients from being counted multiple times, the two patients' appointment dates or appointment IDs can be used to determine if there are duplicate counts; that is, duplicates are removed if the dates or appointment numbers are the same.

[0047] For example, at Hospital A, the first patient on the first working day of 2021 was A, with an appointment time of 08:00:00 on January 4, 2021; the second patient, B, had an appointment time of 09:00:00 on January 4, 2021; the third patient, C, had an appointment time of 09:45:00 on January 4, 2021; the fourth patient, D, had an appointment time of 11:10:00 on January 4, 2021; the fifth patient, E, had an appointment time of 11:20:00 on January 4, 2021; the sixth patient, F, had an appointment time of 13:00:00 on January 4, 2021; the seventh patient, G, had an appointment time of 13:10:00 on January 4, 2021; and the eighth patient, F, had an appointment time of 16:00:00 on January 4, 2021. The first patient at Hospital A on the second working day of 2021 was patient A, whose consultation time was 08:00:00 on January 5, 2021; the second patient was patient B, whose consultation time was 09:00:00 on January 5, 2021.

[0048] When the time window is 2 hours long and moves on to each patient:

[0049] The first window contains A, B, and C. Then AB, BC, and AC are associated.

[0050] The second window consists of B and C, which overlap with the first window and are subsets of it; therefore, they are not counted.

[0051] If the third window contains C and D, then CD will accompany...

[0052] The fourth window is: D, E, F, then DE, DF, EF are associated.

[0053] The fifth window contains E, F, and G. Then EF, EG, and FG are associated with it. Since the dates of EF are 2021-01-04 and 2021-01-04, which are the same as EF in the fourth window, we remove duplicates and keep only EG and FG.

[0054] The 6th window is F, G, which is a subset of the 5th window and is not calculated.

[0055] The 7th window is: H, a patient, not counted.

[0056] The 8th window contains: A and B. Although the AB association was previously calculated, the dates are 2021-01-05 and 2021-01-05, which are different from the dates of the previously occurring associations. Therefore, the AB association count is increased by 1.

[0057] Therefore, in this example, the number of times AB is accompanied by behavior is 2, while the number of times the other accompanying behaviors BC, AC, CD, DE, DF, EF, EG, and FG are all 1.

[0058] After obtaining all medical records from the same medical institution over a period of time and extracting accompanying behaviors, the accompanying behaviors of two patients meeting predetermined conditions can be used to determine that the two patients have an accompanying relationship. In one embodiment, the predetermined conditions may include the number of accompanying behaviors between the two patients reaching a predetermined threshold number. The threshold number may be a threshold number of occurrences within a specific time period, or it may be the total number of occurrences without a specified time period. For example, the threshold number may be set at 5 accompanying behaviors within 3 months; or it may be set at 8 accompanying behaviors without a time limit, etc.

[0059] In the above embodiments, the accompanying behaviors of patients within the same medical institution can be used to determine companionship relationships and identify suspected groups. In other embodiments, the accompanying behaviors of multiple medical institutions can be considered collaboratively to determine companionship relationships. For this purpose, patient visit records from all medical institutions within a specific region can be obtained, and the accompanying behaviors of all medical institutions can be used to determine the companionship relationship; alternatively, patient visit records from medical institutions in a specific area within a specific region can be obtained, and the accompanying behaviors of the medical institutions in that specific area can be used to determine the companionship relationship.

[0060] For example, in some embodiments, medical records from major medical institutions throughout the city can be collected to find pairwise accompanying behaviors of patients within the same medical institution during the same time period, thereby determining the accompanying relationship. In other embodiments, medical records from specialized medical institutions, such as those in the field of pulmonology, can be collected to find pairwise accompanying behaviors of patients within the same medical institution during the same time period, thereby determining the accompanying relationship. In these cross-institutional accompanying relationship determinations, the predetermined conditions may also involve the relationship between the various medical institutions where the accompanying behavior occurred. For example, accompanying behaviors between different medical institutions can reduce the overall threshold requirement. For instance, if patients A and B have 3 accompanying behaviors at Hospital A and 3 accompanying behaviors at Hospital B, a specialized hospital 15 kilometers away from Hospital A, the aforementioned accompanying behaviors can be equivalent to patients A and B having 8 accompanying behaviors only at Hospital A (i.e., the threshold number is reduced due to the existence of cross-hospital behaviors). This is because the probability of patients A and B appearing multiple times at Hospital B simultaneously without prior agreement is very low.

[0061] Furthermore, given the common occurrence of families and friends making appointments to visit each other in real life, especially elderly people with chronic illnesses, the identification method of this invention can also exclude patients with comorbid relationships based on certain conditions to prevent misjudgment. For example, patients whose visit intervals and / or medication records do not meet the abnormal judgment criteria can be excluded. Similarly, for patients with diseases requiring frequent visits, such as those with kidney failure requiring frequent hemodialysis, a diagnostic whitelist can be established in advance based on industry experience in the medical field, and patients can be removed based on this whitelist. Simultaneously, visits with low diagnostic costs or low reimbursement rates can also be used to exclude patients without abnormal behavior.

[0062] After identifying the companion relationships, multiple patients who all share these relationships can be identified from among the various companion relationships, and these patients can be identified as a group engaging in abnormal medical behavior. Here, "each patient sharing a companion relationship" means that every pair of patients in the group has a companion relationship with each other. In one embodiment, graph theory can be used to ensure accurate identification of abnormal behavior groups. For this purpose, patients can be represented as nodes, and companion relationships can be represented as edges to construct a companion network. Maximal clusters in the companion network where every pair of nodes is connected by an edge are identified as groups suspected of engaging in abnormal medical behavior.

[0063] Specifically, when the accompanying behaviors between patients meet predetermined conditions, it can be determined that two patients constitute an accompanying relationship; that is, when patients are considered as nodes, there is an edge connecting these two nodes. Therefore, an accompanying graph, i.e., an undirected accompanying graph G, or accompanying network, can be constructed from the patients with edge connections and the edges between them.

[0064] After obtaining the adjoint undirected graph G, we can calculate the cliques, or maximal cliques. Here, a clique refers to a complete subgraph of an undirected graph G, in which every pair of nodes must be connected by an edge. A clique is called a maximal clique if it is not contained in any other clique, i.e., it is not a proper subset of any other clique.

[0065] If multiple patients constitute a suspected group, then any two members of this suspected group will have a companion relationship, meaning there is a connecting edge between any two nodes. Therefore, a group is a complete subgraph of graph G, i.e., a clique. Each maximal clique in graph G represents all members of a group (at least all members involved in abnormal medical behavior). For this purpose, a maximal clique algorithm can be used, such as the Bron-Kerbosch algorithm, to find all maximal cliques from the companion network constructed above. If a maximal clique has at least two nodes, then this maximal clique can be identified as a group.

[0066] The above describes the abnormal medical behavior identification method and its preferred embodiments according to the present invention. The present invention can extract accompanying behaviors between patients from existing medical records (usually anonymized), further forming accompanying relationships, and constructing an undirected accompanying graph based on these relationships, thus identifying groups from a graph perspective. Therefore, it can automatically extract accompanying behaviors between patients from medical records, form accompanying relationships, construct an accompanying network (graph), and finally identify all groups exhibiting abnormal behavior. The entire process is completed automatically without human intervention, and the identification of group members is accurate and comprehensive.

[0067] The abnormal medical behavior identification scheme of the present invention can also be implemented as an abnormal medical behavior identification device. This device can be implemented by a server of a medical information processing institution, which can access a database containing patient visit records (which may be anonymized visit records) of the medical institution.

[0068] Figure 3 A schematic diagram of an abnormal medical behavior recognition device according to an embodiment of the present invention is shown. The device 300 includes a time stream processing module 310, an accompaniment relationship extraction module 320, and a network construction and search module 330.

[0069] The time stream processing module 310 is used to divide the patient visit records of medical institutions into time segments to obtain multiple visit time slices for the same medical institution.

[0070] The accompanying relationship extraction module 320 is used to extract accompanying behaviors from the consultation time slice and filter out accompanying relationships that meet predetermined conditions from the extracted accompanying behaviors.

[0071] The network construction and search module 330 is used to construct a companion network based on the companion relationship, and search for multiple patients with companion relationships with each other in the companion network as a group suspected of having abnormal medical behavior.

[0072] Furthermore, although not shown in the figure, the device 300 may also include: an abnormal medical expense alarm module, which alarms the medical expense entries of patients in the group when the disease type and reimbursement amount of the group suspected of having abnormal medical behavior are not removed from the preset whitelist.

[0073] The abnormal medical behavior recognition scheme of the present invention can also be implemented as an abnormal medical behavior recognition interactive system. This system can be implemented as a client that can control the above-mentioned abnormal medical behavior recognition device.

[0074] Figure 4 A schematic diagram of an interactive system for recognizing abnormal medical behavior according to an embodiment of the present invention is shown. The interactive system includes a settings interface module 410 and a display interface module 420.

[0075] The settings interface module 410 is used to set the range of patient visit records to be selected from the database and to set the duration of each visit time slice. The display interface module 420 is used to display a companion network consisting of accompanying behaviors extracted from the visit time slices that meet predetermined conditions, and to highlight multiple patient nodes in the companion network that have companion relationships with each other.

[0076] Furthermore, although not shown in the figure, the display interface module 420 also includes an operable module for displaying the medical consumption records of the patient node when selected.

[0077] Specifically, users can open a client application that connects to and controls the server. In the client's settings interface, users can set the range of medical records to be selected (e.g., which medical institutions to select, the time span of medical records to select (e.g., the past year), etc.), and also set the length and granularity of the medical time slices (e.g., a duration of 2 hours, with time windows sliding at the individual patient level). After setting, the backend performs calculations according to the identification method of the present invention and presents the results on a results display interface. This results display interface can show the accompanying behaviors between patients in a graphical form and can highlight suspected groups. Users can, for example, click on the corresponding node to view the medical insurance consumption records of members of the suspected group and make corresponding judgments.

[0078] Figure 5A schematic diagram of a computing device that can be used to implement the above-described abnormal medical behavior identification method according to an embodiment of the present invention is shown.

[0079] See Figure 5 The computing device 500 includes a memory 510 and a processor 520.

[0080] Processor 520 may be a multi-core processor or may contain multiple processors. In some embodiments, processor 520 may include a general-purpose main processor and one or more special-purpose coprocessors, such as a graphics processing unit (GPU), a digital signal processor (DSP), etc. In some embodiments, processor 520 may be implemented using custom circuitry, such as an application-specific integrated circuit (ASIC) or a field-programmable gate array (FPGA).

[0081] Memory 510 may include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage devices. ROM may store static data or instructions required by the processor 520 or other modules of the computer. Permanent storage devices may be read-write storage devices. Permanent storage devices may be non-volatile storage devices that retain stored instructions and data even when the computer is powered off. In some embodiments, permanent storage devices use mass storage devices (e.g., magnetic or optical disks, flash memory) as permanent storage devices. In other embodiments, permanent storage devices may be removable storage devices (e.g., floppy disks, optical drives). System memory may be a read-write storage device or a volatile read-write storage device, such as dynamic random access memory. System memory may store some or all of the instructions and data required by the processor during operation. Furthermore, memory 510 may include any combination of computer-readable storage media, including various types of semiconductor memory chips (DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), and disks and / or optical disks may also be used. In some embodiments, memory 510 may include a removable storage device that is readable and / or writable, such as a laser disc (CD), a read-only digital multifunction optical disc (e.g., DVD-ROM, dual-layer DVD-ROM), a read-only Blu-ray disc, an ultra-high-density optical disc, a flash memory card (e.g., SD card, mini SD card, Micro-SD card, etc.), a magnetic floppy disk, etc. Computer-readable storage media do not contain carrier waves or transient electronic signals transmitted wirelessly or via wired connections.

[0082] The memory 510 stores executable code, which, when processed by the processor 520, enables the processor 520 to execute the abnormal medical behavior recognition method described above.

[0083] Application examples

[0084] To facilitate understanding of the preferred embodiments of the present invention, Figure 6 A schematic diagram illustrating the specific implementation steps of an abnormal medical behavior identification method according to a preferred embodiment of the present invention is shown.

[0085] Since the identification scheme of this invention constructs an accompanying network based on accompanying medical visit behavior, and the definition of an accompanying person is a person who visits the same medical institution within the same time period, it is necessary to determine which medical visits occur within the same time period.

[0086] 1) Time Stream Processing

[0087] All visits to the same medical institution are processed according to a timeline. The medical records for each institution are sorted by visit time. Then, a sliding window is set (the window size is an input parameter h, such as 2 hours). The sliding window starts from the smallest visit time and slides over a span of h time points, sliding one time point at a time (e.g., one patient node). Thus, for the medical data of a medical institution on a certain day, n time windows are obtained through this sliding process.

[0088] 2) Accompanying Behavior Extraction

[0089] For all patients whose medical records appear in the same window, there is one accompanying behavior between each pair of them. Since a soft window is used here (meaning windows may overlap), some patients may have the same medical record appearing in multiple windows, leading to duplicate calculations, such as some overlapping medical records between time window 1 and time window 2. Therefore, to address this issue, when calculating the accompanying behavior of two patients, their corresponding dates (accurate to the day) are used for deduplication; that is, two patients on the same day have only one accompanying behavior. In another embodiment, the medical record ID can also be used for deduplication, which can identify multiple medical visits on the same day.

[0090] 3) Construction of the companion graph

[0091] Since the occurrence of one or a few instances of shared behavior between two patients may be coincidental, if such occurrences are numerous, such as at least five, then a group relationship between them is highly probable. Therefore, this scheme provides a threshold t as a criterion. If the number of times two patients share shared behavior is at least t, then they constitute a shared relationship, i.e., there is an edge connection between them. An shared graph, i.e., a shared undirected graph G, or shared network, is constructed by combining the patients with edge connections with the edges between them.

[0092] 4) Gang Identification

[0093] After obtaining the adjoining undirected graph G, if a suspected community exists, then any two members within that community will have an adjoining relationship, meaning there is a connecting edge between any two nodes. Therefore, a suspected community is a complete subgraph of graph G, i.e., a clique. Thus, each maximal clique in graph G is a suspected community. Therefore, the Bron-Kerbosch maximal clique algorithm can be used to obtain all maximal cliques. If a maximal clique has at least two nodes, then it is a clique. As shown in the figure, three suspected communities with four, five, and three members respectively can be derived from the graph.

[0094] Therefore, the present invention can extract the accompanying behaviors between patients from massive medical data, further form accompanying relationships, construct an accompanying undirected graph based on the accompanying relationships between patients, and identify suspected groups from the perspective of graph.

[0095] The abnormal medical behavior identification method according to the present invention and the computational identification capable of performing the above method have been described in detail above with reference to the accompanying drawings. The solution of the present invention and its preferred embodiments can achieve the following effects:

[0096] 1) Fully automated

[0097] The system automatically extracts accompanying behaviors between patients from their medical records, further forming accompanying relationships to construct an accompanying network (graph), and finally identifies all suspected groups. The entire process is automated and requires no human intervention.

[0098] 2) Precise calculation of accompanying behaviors

[0099] Multiple members of a group often visit the same medical institution (hospital visit or pharmacy purchase) within the same time period. This time period is typically short, such as within two hours, and may span between calendar hours, for example, member A visits at 14:58 and member B at 15:12. If we use a coarse-grained division based on calendar days, there will be omissions of accompanying behaviors (visits that are close together but span different calendar days), and the computational workload will be large (for example, if a tertiary hospital sees 2000 patients a day, then accompanying behaviors need to be calculated for every pair of these 2000 people). If we use a fine-grained division based on calendar hours, there will still be omissions of accompanying behaviors (visits that are close together but span different calendar hours).

[0100] In a preferred embodiment of this solution, accompanying behaviors are identified by using overlapping sliding windows of a specified size for visits to the same medical institution according to the time flow, and accompanying behaviors are counted only once for the same date (or a pair of visit records) to avoid duplicate counting, making the calculation of accompanying behaviors more accurate.

[0101] 3) Identify all groups exhibiting abnormal behavior

[0102] After constructing the adjoint network, the maximal clique identification algorithm can identify all groups with suspected abnormal medical behavior without errors or omissions.

[0103] The abnormal medical behavior identification method of this invention not only abstracts data into entities and relationships but also significantly improves the efficiency of batch correlation calculations, enabling direct processing of massive amounts of data. First, this invention understands data from a correlation perspective, which better aligns with practical needs and allows for the unified combination and processing of various data types (including hospital visits and pharmacy purchases). Second, in the preferred embodiment, a graph perspective is used, abstracting patients as nodes and multiple accompanying behaviors between patients as edges, thereby improving interpretability and intuitiveness. Third, the identification scheme of this invention features more precise time processing, ensuring no accompanying behaviors are missed, thus identifying all accompanying relationships. Furthermore, the maximal clique algorithm ensures the identification of all suspected groups and all members of each group, further avoiding the omission of suspected groups or group members. Finally, this solution can be fully automated, greatly facilitating audits.

[0104] Furthermore, the method according to the present invention can also be implemented as a computer program or computer program product, which includes computer program code instructions for performing the steps defined in the above-described method of the present invention.

[0105] Alternatively, the present invention can also be implemented as a non-transitory machine-readable storage medium (or computer-readable storage medium, or machine-readable storage medium) storing executable code (or computer program, or computer instruction code) thereon, which, when executed by a processor of an electronic device (or computing device, server, etc.), causes the processor to perform the various steps of the method described above according to the present invention.

[0106] Those skilled in the art will also understand that the various exemplary logic blocks, modules, circuits, and algorithm steps described in connection with the disclosure herein can be implemented as electronic hardware, computer software, or a combination of both.

[0107] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems and methods according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0108] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for identifying abnormal medical behavior, comprising: Divide patient visit records at medical institutions into time segments to obtain multiple visit time slices for the same medical institution; The accompanying behaviors are extracted from the consultation time slices, and the accompanying relationships that meet the predetermined conditions are selected from the extracted accompanying behaviors. as well as Represent patients as nodes and adjoint relationships as edges to construct an adjoint network; A maximal clique is identified in the adjoint network that is connected by an edge between each pair of nodes and contains at least two nodes, as a group suspected of having abnormal medical behavior. Extracting accompanying behaviors from the consultation time slices, and filtering out accompanying relationships that meet predetermined conditions from the extracted accompanying behaviors, includes: All patients within the same consultation time frame are considered to have a single accompanying behavior in each pair; as well as If the number of times the accompanying behaviors between two patients reaches a predetermined threshold, it is determined that the two patients have an accompanying relationship.

2. The method as described in claim 1, wherein, Divide patient visit records at medical institutions into time slices to obtain multiple visit time slices for the same medical institution, including: Sort patient visit records from a specific medical institution on a specific date in chronological order; and The medical records are divided into medical time slices using a time window that is equal to the predetermined duration and slides according to the time flow.

3. The method as described in claim 1, wherein, The method includes at least one of the following: To obtain patient medical records from a specific medical institution. Furthermore, the accompanying relationship is determined solely by extracting accompanying behaviors from the patient's visit time slices at the medical institution; or Obtain patient medical records from all medical institutions within a specific region. Furthermore, the accompanying behaviors are extracted from the visit time slices of all the medical institutions to determine the accompanying relationship; or To obtain patient medical records from medical institutions in a specific region and field. Furthermore, the accompanying behaviors are extracted from the consultation time slices of the medical institutions in the specific field to determine the accompanying relationship.

4. The method of claim 1, wherein, Individuals engaging in normal medical practices are removed from groups suspected of having abnormal medical practices, based on the type of disease and the amount of money spent.

5. An abnormal medical behavior recognition device, comprising: The time stream processing module is used to divide the patient visit records of medical institutions into time segments to obtain multiple visit time slices for the same medical institution. The accompanying relationship extraction module is used to extract accompanying behaviors from the consultation time slice and filter out accompanying relationships that meet predetermined conditions from the extracted accompanying behaviors. as well as The network construction and search module is used to represent patients as nodes and companion relationships as edges to construct a companion network; and to identify maximal cliques in the companion network that are connected by edges between each pair of nodes and contain at least two nodes as suspected groups of abnormal medical behavior. The accompanying relationship extraction module is specifically used to determine that all patients within the same consultation time slice have a pair of accompanying behaviors; The relationship between two patients is determined by the number of times the number of times the accompanying behaviors between the two patients reach a predetermined threshold.

6. The apparatus of claim 5, further comprising: The abnormal medical expense alarm module will issue an alarm for the medical expense items of patients in a group that is suspected of having abnormal medical behavior, when the disease type and reimbursement amount of the group have not been removed from the preset whitelist.

7. An abnormal medical behavior recognition and interaction system, comprising: The settings interface module is used to set the range of patient medical records to be selected from the database for processing and to set the duration of the medical time slice. The display interface module is used to display a companion network consisting of companion behaviors that meet predetermined conditions extracted from the consultation time slice, and to highlight multiple patient nodes in the companion network that have companion relationships with each other. The display interface includes an operable module for displaying the medical consumption records of the patient node when it is selected. All patients within the same consultation time slot exhibit a pairwise accompanying behavior once. When the number of accompanying behaviors between two patients reaches a predetermined threshold, the two patients are considered to have an accompanying relationship. The accompanying network is a graph structure constructed by representing patients as nodes and accompanying relationships as edges. In the accompanying network, each pair of nodes is connected by an edge, and a maximal clique containing at least two nodes is identified as a group suspected of having abnormal medical behavior.

8. A non-transitory machine-readable storage medium having executable code stored thereon, which, when executed by a processor of an electronic device, causes the processor to perform the method as described in any one of claims 1-4.

Images