A data analysis and evaluation method for medical insurance violations

Abstract

The present application relates to the field of data processing, especially to a medical insurance illegal behavior data analysis and evaluation method, the method comprises the following steps: collecting and preprocessing the medical insurance outpatient settlement flow data to obtain medical insurance outpatient settlement characteristic data; based on the spatial hidden camouflage factor of cross-medical institution drug acquisition distribution characteristics, and based on the time and space folding compression factor obtained by the sequence of drug purchase time stamp and the sequence of available days of prescription regulations; the nonlinear correction of the basic Mahalanobis distance is combined with the spatial hidden camouflage factor and the time and space folding compression factor to obtain the reconstructed Mahalanobis distance, and the evaluation result of medical insurance illegal behavior is obtained by comparing with the abnormal judgment threshold, so as to improve the identification accuracy of hidden illegal behaviors such as cross-institutional scattered drug purchase and short-time repeated drug purchase.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a data analysis and evaluation method for medical insurance violations. Background Technology

[0002] With the continuous advancement of medical security informatization, the medical insurance settlement system is accumulating a large amount of outpatient chronic disease medication purchase data. To enhance the analysis of medical insurance fund usage behavior, existing technologies typically construct anomaly detection models based on aggregated characteristics such as the total number of visits, total number of prescriptions, and total settlement amount within a preset period. They also employ statistical distance metrics such as Mahalanobis distance to assess the degree of deviation of patients' medication behavior from the distribution center of the normal population, in order to identify potentially abnormal medication purchase behaviors.

[0003] However, in actual medical insurance settlement scenarios, some abnormal drug purchase behaviors reduce the amount of medication purchased per transaction by obtaining the same drug from multiple medical institutions within a short period of time. This keeps the total number of visits, total number of prescriptions, and total settlement amount within the normal range. Existing feature aggregation methods based on scalar accumulation mainly reflect changes in the total amount of data, making it difficult to retain the distribution differences in the quantity of medication obtained across different medical institutions, and also difficult to characterize the overlap of the prescribed number of days of use for prescriptions corresponding to adjacent drug purchase records on the actual time axis. As a result, features such as spatial dispersion and temporal folding are weakened or lost during the aggregation process.

[0004] Therefore, when overall characteristics are similar, anomalous samples formed by drug purchases across different medical institutions are prone to resemble normal patient groups in the feature space, leading to a lower baseline Mahalanobis distance and false negatives. How to further extract the distribution characteristics of drug acquisition across medical institutions and the overlap characteristics of prescription time while retaining the existing statistical anomaly detection framework, and how to use spatiotemporal correlation information to correct the anomaly distance assessment results, has become an urgent problem to be solved in the field of data analysis and evaluation of medical insurance violations. Summary of the Invention

[0005] In view of this, the present invention aims to propose a data analysis and evaluation method for medical insurance violations, in order to solve the problem that existing methods are unable to identify hidden violations caused by scattered drug purchases across institutions and overlapping prescription times.

[0006] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0007] A data analysis and evaluation method for medical insurance violations, the method comprising:

[0008] Step S1: Collect and preprocess medical insurance outpatient settlement transaction data to obtain medical insurance outpatient settlement characteristic data;

[0009] Step S2: Obtain the spatial concealment and camouflage factor by performing decentralized mapping processing on the cross-medical institution drug acquisition distribution characteristics in the medical insurance outpatient settlement characteristic data;

[0010] Step S3: Obtain the spatiotemporal folding compression factor by performing time folding evaluation on the drug purchase timestamp sequence and the prescription usable days sequence in the medical insurance outpatient settlement feature data;

[0011] Step S4: Obtain the basic Mahalanobis distance by calculating the Mahalanobis distance of the global patient basic feature matrix, and obtain the reconstructed Mahalanobis distance by performing nonlinear joint correction on the basic Mahalanobis distance, spatial concealment camouflage factor and spatiotemporal folding compression factor.

[0012] Step S5: By comparing and analyzing the reconstructed Mahalanobis distance with the anomaly judgment threshold, the assessment results of medical insurance violations are obtained.

[0013] Furthermore, the process of collecting and preprocessing medical insurance outpatient settlement transaction data to obtain medical insurance outpatient settlement characteristic data includes:

[0014] Obtain the outpatient chronic disease medication purchase settlement data recorded by the medical insurance settlement system within the target area during the preset evaluation period and perform data anonymization; for any outpatient chronic disease medication purchase settlement record, extract the patient's unique identification code, medical institution code, medication purchase timestamp, unified drug code, drug quantity, and settlement amount, and use the extracted data as medical insurance outpatient settlement data;

[0015] Using the patient's unique identification code as the data aggregation dimension, abnormal records are removed and the time sequence is processed in the medical insurance outpatient settlement data. Outpatient chronic disease drug purchase settlement records with empty drug codes or abnormal drug prescription quantities are removed. Based on the drug instructions or the daily standard dosage of the corresponding drugs in the medical insurance catalog, the drug prescription quantities in each outpatient chronic disease drug purchase settlement record are converted into the corresponding prescription usage days. Outpatient chronic disease drug purchase settlement records for the same patient are arranged in chronological order of purchase timestamps.

[0016] For any target patient, the total number of visits, total number of prescriptions, and total settlement amount are statistically obtained based on the outpatient chronic disease medication purchase settlement records within the preset assessment period. The total number of visits, total number of prescriptions, and total settlement amount are used as the first type of summary feature of the target patient. For the target drug, the total number of different medical institutions visited by the target patient, the cumulative total quantity of the target drug, and the maximum quantity of the target drug prescribed by a single medical institution are statistically obtained. At the same time, the medication purchase timestamp sequence and the prescription validity period sequence corresponding to the target patient are obtained. The total number of different medical institutions, the cumulative total quantity of the target drug, the maximum quantity of the target drug prescribed by a single medical institution, the medication purchase timestamp sequence, and the prescription validity period sequence are used as the second type of spatiotemporal trajectory feature of the target patient.

[0017] The first type of aggregated features of all patients within the preset assessment period are combined according to the patient dimension to obtain the global patient basic feature matrix. The global patient basic feature matrix and the second type of spatiotemporal trajectory features corresponding to all patients are used as medical insurance outpatient settlement feature data.

[0018] Furthermore, the process of obtaining spatial concealment and camouflage factors by performing decentralized mapping processing on the cross-medical institution drug acquisition distribution characteristics in the medical insurance outpatient settlement feature data includes:

[0019] By performing cross-institutional distribution feature extraction on the total number of medical institutions visited and the distribution of drug acquisition quantity in the medical insurance outpatient settlement feature data, institutional divergence constraint data and drug decentralization depth data are obtained.

[0020] By cross-coupling and exponential mapping of institutional divergence constraint data and decentralized drug depth data, spatial concealment and camouflage factors are obtained.

[0021] Furthermore, the process of extracting cross-institutional distribution features from the total number of medical institutions visited and the distribution data of drug acquisition quantities in the medical insurance outpatient settlement feature data to obtain institutional divergence constraint data and drug decentralization depth data includes:

[0022] For any target patient, extract the total number of medical institutions visited by the target patient, the total number of target drugs, and the maximum number of target drugs prescribed by a single medical institution from the medical insurance outpatient settlement feature data.

[0023] The difference between the total number of medical institutions visited by the target patient and the constant 1 is used as the evaluation of the number of institutions dispersed for the target patient. The evaluation of the number of institutions dispersed is used as the numerator, the total number of medical institutions visited is used as the denominator, and the resulting fraction is used as the institutional divergence constraint data for the target patient.

[0024] The difference between the total cumulative quantity of target drugs for target patients and the maximum quantity of target drugs prescribed by a single medical institution is used as the drug quantity assessment for non-primary medical institutions for target patients. The drug quantity assessment for non-primary medical institutions is used as the numerator, and the total cumulative quantity of target drugs is used as the denominator. The resulting fraction is used as the decentralized deep data of drugs for target patients.

[0025] Furthermore, the process of obtaining spatial concealment and camouflage factors by cross-coupling and exponentially mapping institutional divergence constraint data with decentralized drug depth data includes:

[0026] For any target patient, obtain the institutional divergence constraint data and the drug decentralized depth data corresponding to the target patient;

[0027] Multiply the institutional divergence constraint data corresponding to the target patient with the decentralized depth data of the drug, and use the resulting product as the cross-institutional decentralized cross-evaluation corresponding to the target patient.

[0028] An exponential mapping with a natural constant as the base is performed on the cross-institutional decentralized cross-assessment corresponding to the target patient, and the corresponding mapping result is used as the spatial concealment camouflage factor corresponding to the target patient.

[0029] Furthermore, the step of obtaining a spatiotemporal folding compression factor by performing time folding evaluation on the drug purchase timestamp sequence and the prescription usable days sequence in the medical insurance outpatient settlement feature data includes:

[0030] By performing adjacent time-series correlation processing on the drug purchase timestamp sequence and the prescription usable days sequence in the medical insurance outpatient settlement feature data, adjacent drug purchase time interval data and corresponding prescription usable duration data are obtained.

[0031] By performing time overlap truncation and accumulation processing on adjacent drug purchase time interval data and corresponding prescription availability time data, cumulative prescription time overlap data is obtained.

[0032] By normalizing and mapping the cumulative prescription time overlap data with the drug purchase time span data corresponding to the drug purchase timestamp sequence, the spatiotemporal folding compression factor is obtained.

[0033] Furthermore, the step of performing adjacent time-series correlation processing on the drug purchase timestamp sequence and the prescription availability days sequence in the medical insurance outpatient settlement feature data to obtain adjacent drug purchase time interval data and corresponding prescription availability duration data includes:

[0034] For any target patient, extract the drug purchase timestamp sequence and prescription validity period sequence corresponding to the target patient from the medical insurance outpatient settlement feature data, and count the total number of drug purchase records corresponding to the target patient within the preset evaluation period;

[0035] For any target drug purchase record other than the first drug purchase record in the drug purchase timestamp sequence corresponding to the target patient, obtain the drug purchase timestamp corresponding to the target drug purchase record and the drug purchase timestamp corresponding to the previous drug purchase record. Use the difference between the drug purchase timestamp corresponding to the target drug purchase record and the drug purchase timestamp corresponding to the previous drug purchase record as the adjacent drug purchase time interval data corresponding to the target drug purchase record.

[0036] Extract the number of days the prescription is available from the sequence of days the prescription is available for the target patient, and use the number of days the prescription is available for the previous purchase record as the prescription availability duration data for the target purchase record.

[0037] Furthermore, the step of obtaining cumulative prescription time overlap data by performing time overlap truncation and accumulation processing on adjacent drug purchase time interval data and corresponding prescription availability time data includes:

[0038] For any target patient and any target medication purchase record other than the first medication purchase record, obtain the adjacent medication purchase time interval data and prescription availability duration data corresponding to the target medication purchase record;

[0039] The difference between the available prescription duration data corresponding to the target drug purchase record and the adjacent drug purchase time interval data is used as the initial prescription time overlap assessment for the target drug purchase record; when the initial prescription time overlap assessment is greater than the constant 0, the initial prescription time overlap assessment is used as the prescription time overlap assessment for the target drug purchase record; when the initial prescription time overlap assessment is less than or equal to the constant 0, the prescription time overlap assessment for the target drug purchase record is set to the constant 0.

[0040] The prescription time overlap assessments for all medication purchase records (excluding the first purchase record) corresponding to the target patient are summed up, and the resulting sum is used as the cumulative prescription time overlap data for the target patient.

[0041] Furthermore, the step of obtaining the spatiotemporal folding compression factor by normalizing and mapping the cumulative prescription time overlap data with the purchase time span data corresponding to the purchase timestamp sequence includes:

[0042] For any target patient, extract the drug purchase timestamp sequence corresponding to the target patient from the medical insurance outpatient settlement feature data, and obtain the drug purchase timestamp corresponding to the first drug purchase record and the drug purchase timestamp corresponding to the last drug purchase record in the drug purchase timestamp sequence.

[0043] The difference between the timestamp of the last medication purchase record and the timestamp of the first medication purchase record is used as the initial medication purchase time span assessment for the target patient. When the initial medication purchase time span assessment is greater than or equal to a constant 1, the initial medication purchase time span assessment is used as the medication purchase time span data for the target patient. When the initial medication purchase time span assessment is less than a constant 1, the medication purchase time span data for the target patient is set to a constant 1.

[0044] Obtain the cumulative prescription time overlap data corresponding to the target patient, and use the calculation result of adding the cumulative prescription time overlap data and the drug purchase time span data as the time folding normalization scale corresponding to the target patient; use the cumulative prescription time overlap data as the numerator, the time folding normalization scale as the denominator, and use the corresponding fraction as the spatiotemporal folding compression factor corresponding to the target patient.

[0045] Furthermore, the step of obtaining the basic Mahalanobis distance by calculating the Mahalanobis distance on the global patient baseline feature matrix, and obtaining the reconstructed Mahalanobis distance by performing nonlinear joint correction on the basic Mahalanobis distance, the spatial concealment camouflage factor, and the spatiotemporal folding compression factor, includes:

[0046] Based on the global patient basic feature matrix, the mean of the total number of visits, total number of prescriptions and total settlement amount for all patients within the preset assessment period is calculated to obtain the global mean feature vector. Based on the global patient basic feature matrix, the variance and covariance between the basic features of all patients are calculated to obtain the global covariance matrix and the corresponding global covariance inverse matrix.

[0047] For any target patient, extract the first type of summary features corresponding to the target patient from the global patient basic feature matrix, and construct the target patient basic feature vector based on the total number of visits, total number of prescriptions, and total settlement amount corresponding to the target patient; take the difference between the target patient basic feature vector and the global mean feature vector as the target patient's feature deviation vector, perform matrix multiplication on the transpose of the feature deviation vector, the global covariance inverse matrix, and the feature deviation vector in sequence, take the corresponding calculation result as the squared Mahalanobis distance evaluation corresponding to the target patient, and take the square root of the squared Mahalanobis distance evaluation as the basic Mahalanobis distance corresponding to the target patient;

[0048] Obtain the spatial concealment and camouflage factor and the spatiotemporal folding compression factor corresponding to the target patient. Use the cube of the spatiotemporal folding compression factor as the time folding nonlinear assessment corresponding to the target patient. Multiply the spatial concealment and camouflage factor and the time folding nonlinear assessment. Use the corresponding product as the spatiotemporal camouflage joint assessment corresponding to the target patient.

[0049] The spatiotemporal camouflage joint assessment corresponding to the target patient is subjected to an exponential mapping with the natural constant as the base, and the corresponding mapping result is used as the distance reconstruction amplification factor corresponding to the target patient; the basic Mahalanobis distance corresponding to the target patient is multiplied by the distance reconstruction amplification factor, and the corresponding product is used as the reconstructed Mahalanobis distance corresponding to the target patient.

[0050] Compared with the prior art, the present invention has the following advantages:

[0051] This invention provides a data analysis and evaluation method for medical insurance violations. By jointly characterizing the dispersion of institutional distribution, the degree of decentralization of drug quantity, and the overlap of prescription availability time on the actual drug purchase timeline when patients obtain similar drugs from multiple medical institutions, it can re-extract the spatiotemporal correlation information that is compressed and lost in traditional total aggregation features. For concealed drug purchase behaviors such as visiting multiple medical institutions in a short period of time, obtaining small amounts of medication at a time but with abnormal cumulative acquisition, this method can effectively amplify the data differences between these behaviors and those of normal patients who have stable medical visits and follow-up visits according to the actual medication cycle, thereby reducing false negatives caused by fragmented drug purchases across institutions. At the same time, this invention uses spatiotemporal folding features to constrain the cross-institutional distribution features and reconstructs the basic Mahalanobis distance through a nonlinear joint correction mechanism. This suppresses minor abnormal features caused by normal out-of-town medical visits, travel-related drug purchases, and reasonable advance drug preparation to a certain extent, while significantly enhancing abnormal behaviors that simultaneously exhibit cross-institutional decentralized distribution and high-intensity prescription time overlap. This will improve the ability and accuracy of medical insurance outpatient settlement anomaly assessment in distinguishing and judging hidden and illegal drug purchase behaviors while retaining the existing Mahalanobis distance statistical discrimination framework, and provide more reliable data basis for medical insurance fund auditors to conduct high-risk sample screening and subsequent verification. Attached Figure Description

[0052] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0053] Figure 1 This is a flowchart illustrating a data analysis and evaluation method for medical insurance violations as described in an embodiment of the present invention. Detailed Implementation

[0054] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0055] See Figure 1 This is a flowchart of a data analysis and evaluation method for medical insurance violations provided in Embodiment 1 of the present invention. Figure 1 As shown, a data analysis and evaluation method for medical insurance violations may include:

[0056] Step S1: Collect and preprocess medical insurance outpatient settlement transaction data to obtain medical insurance outpatient settlement characteristic data.

[0057] First, obtain the outpatient chronic disease medication purchase settlement data recorded by the medical insurance settlement system within the target area during the preset evaluation period and perform data anonymization. For any outpatient chronic disease medication purchase settlement record, extract the patient's unique identification code, medical institution code, medication purchase timestamp, unified drug code, medication quantity, and settlement amount. Use the extracted data as the medical insurance outpatient settlement data.

[0058] Using the patient's unique identification code as the data aggregation dimension, abnormal records were removed and the data was processed according to time sequence for outpatient medical insurance settlement records. Outpatient chronic disease medication purchase settlement records with empty drug codes or abnormal drug prescription quantities were removed. Based on the daily standard dosage of the corresponding drugs in the drug instructions or the medical insurance catalog, the drug prescription quantities in each outpatient chronic disease medication purchase settlement record were converted into the corresponding prescription's allowed number of days of use. Outpatient chronic disease medication purchase settlement records for the same patient were then arranged in chronological order of purchase timestamps.

[0059] For any target patient, the total number of visits, total number of prescriptions, and total settlement amount are statistically obtained based on the outpatient chronic disease medication purchase settlement records within a preset assessment period. These are used as the first type of summary feature for the target patient. In this embodiment, the assessment period is set to 90 days, a parameter that can be adjusted based on the average follow-up visit cycle and prescription cycle for outpatient chronic disease patients within the target coverage area. For the target medication, the total number of different medical institutions visited by the target patient, the cumulative total quantity of the target medication, and the maximum quantity of the target medication prescribed by a single medical institution are statistically obtained. Simultaneously, the medication purchase timestamp sequence and the prescription's allowed days sequence for the target patient are obtained. These are used as the second type of spatiotemporal trajectory feature for the target patient. The first type of summary features for all patients within the preset assessment period are combined according to the patient dimension to obtain a global patient basic feature matrix. This global patient basic feature matrix and the second type of spatiotemporal trajectory features for all patients are used as medical insurance outpatient settlement feature data.

[0060] This completes the process of collecting and preprocessing medical insurance outpatient settlement transaction data to obtain characteristic data of medical insurance outpatient settlement.

[0061] Step S2 involves obtaining a spatial concealment factor by performing decentralized mapping processing on the cross-medical institution drug acquisition distribution characteristics in the medical insurance outpatient settlement characteristic data.

[0062] Existing medical insurance risk control assessment models based on Mahalanobis distance use a scalar accumulation method to summarize various medication data within a period when extracting patient data features. This aggregation operation logically assumes that the total quantity of medications acquired is the core indicator for measuring violation risk, without imposing feature constraints on the institutional distribution structure of medication acquisition locations. However, in actual medical insurance settlement data, even normal patients who need to purchase medication multiple times for follow-up visits typically have a highly dependent primary medical institution, and their prescription volume shows a clear primary-secondary concentration in spatial institutional distribution. Violators, in order to circumvent the system's single-institution quota interception, must adopt an averaging strategy, resulting in a fragmented distribution of their medication purchases across different medical institutions. The existing scalar accumulation structure cannot identify this difference in data distribution patterns between institutions. During feature aggregation, it treats highly concentrated normal medication use and deliberately averaging illegal medication use as equivalent states, thus masking the deliberately hidden institutional averaging characteristics of violators at the data extraction level. This prevents the assessment model from effectively distinguishing behavioral patterns with the same total amount but different spatial distribution motivations. To address the issue of spatial distribution feature loss caused by aggregated totals, a cross-institutional distribution depth mapping mechanism needs to be introduced into the evaluation model. This mechanism utilizes the individual data distribution boundaries of each institution within a period to construct a numerical feature that remains insensitive to data with clearly defined primary and secondary characteristics but exhibits a nonlinear amplification response to decentralized, evenly distributed data. By structurally penalizing fragmented concealment behavior along the feature dimension, it compensates for the spatial data feature loss caused by the basic algorithm when processing disguised drug purchase fragments.

[0063] In summary, this invention first extracts cross-institutional distribution features from the total number of medical institutions visited and the distribution of drug purchases in the medical insurance outpatient settlement feature data to obtain institutional divergence constraint data and drug decentralization depth data. Specifically, for any target patient, the total number of medical institutions visited, the cumulative total number of target drugs, and the maximum number of target drugs prescribed by a single medical institution are extracted from the medical insurance outpatient settlement feature data. The difference between the total number of medical institutions visited and a constant 1 is used as the institutional dispersion quantity assessment for the target patient. This institutional dispersion quantity assessment is used as the numerator, and the total number of medical institutions visited is used as the denominator. The resulting fraction is used as the institutional divergence constraint data for the target patient.

[0064] The difference between the total cumulative quantity of target drugs for target patients and the maximum quantity of target drugs prescribed by a single medical institution is used as the drug quantity assessment for non-primary medical institutions for target patients. The drug quantity assessment for non-primary medical institutions is used as the numerator, and the total cumulative quantity of target drugs is used as the denominator. The resulting fraction is used as the decentralized deep data of drugs for target patients.

[0065] After obtaining the institutional divergence constraint data and the decentralized drug depth data, the spatial concealment camouflage factor is obtained by cross-coupling and exponential mapping of these two data. Specifically, for any target patient, the institutional divergence constraint data and the decentralized drug depth data corresponding to that patient are obtained. The institutional divergence constraint data and the decentralized drug depth data corresponding to the target patient are multiplied, and the resulting product is used as the cross-institutional decentralized cross-evaluation corresponding to the target patient. An exponential mapping with a natural constant as the base is performed on the cross-institutional decentralized cross-evaluation corresponding to the target patient, and the resulting mapping is used as the spatial concealment camouflage factor corresponding to the target patient.

[0066] In one implementation, it is assumed that the total number of medical institutions visited by the target patient is The cumulative total number of target drugs corresponding to the target patients is: The maximum number of target drugs that a single medical institution can prescribe for a target patient is [number missing]. Then, the calculation expression for the spatial concealment and camouflage factor corresponding to the target patient is:

[0067]

[0068] in, This indicates the spatial concealment and camouflage factor corresponding to the target patient; This represents an exponential function with the natural constant e as the base. This represents the total number of medical institutions visited by the target patient. This represents the total cumulative quantity of the target drugs corresponding to the target patients; This indicates the maximum quantity of the target drug that a single medical institution can prescribe for the target patient.

[0069] It should be noted that, to address the issue that the total aggregation method cannot identify cross-institutional spoofing behavior, this embodiment constructs a cross-penalty product structure consisting of an institutional divergence constraint term and a decentralized depth term. Among these, the constraint term... The decentralized depth term is calculated by determining the discrete baseline based on the number of institutions. The degree of data fragmentation is captured by calculating the proportion of medications taken at non-primary care institutions relative to the total medication dosage. In actual data processing, if a patient's medication behavior corresponds to a normal single-institution visit... The constraint term will be strictly equal to 0, thus forcibly clearing the cross product to zero. If a patient visits multiple institutions but their medication purchases are concentrated at their primary institution (i.e.,...) near The decentralization depth term approaches 0, resulting in an extremely low overall cross-product. This cross-product design based on data extremum constraints ensures the algorithm maintains a high tolerance for genuine medication use behaviors with primary and secondary medical institution dependencies. Simultaneously, when dealing with violations involving the splitting of prescriptions across multiple institutions, both institutional divergence and decentralization depth reach high values, effectively activating the internal product structure. Furthermore, since existing linear distance metrics cannot highlight such camouflage features in multidimensional space, this embodiment introduces an exponential function on the outer side. This exponential structure applies a non-linear numerical amplification mapping to the internal cross-product as it increases, transforming the decentralized averaging behavior originally hidden under the total quantity feature into a penalty weight response significantly higher than the normal range. This process relies entirely on the data boundary extremum differences of the patient's own medication purchase trajectory, achieving feature extraction of spatially fragmented camouflage features.

[0070] Thus, the spatial concealment and camouflage factor was obtained by decentralizing the distribution characteristics of drug acquisition across medical institutions in the medical insurance outpatient settlement feature data.

[0071] Step S3: The spatiotemporal folding compression factor is obtained by performing time folding evaluation on the drug purchase timestamp sequence and the prescription usable days sequence in the medical insurance outpatient settlement feature data.

[0072] After constructing the spatial concealment factor in step S2, the model has the ability to identify the fragmented distribution of drug purchase trajectories among medical institutions. However, when this spatial concealment feature is directly applied to anomaly detection, it faces the problem of data misjudgment interference from normal cross-regional personnel, such as business travelers, commuters, or patients relocated to other places. In actual insurance settlement business, due to the actual geographical location change, the normal drug purchase behavior of cross-regional personnel will also cross two or more different medical institutions, resulting in a certain degree of decentralization in the spatial distribution of their data. At this time, the spatial concealment factor will be activated and output a large feature value. However, there is a core behavioral difference between normal cross-institutional drug purchase and abnormal drug purchase in the time dimension. The motivation of normal patients to purchase drugs is based on the actual consumption of drugs, and the objective time interval between two adjacent purchases is usually greater than or equal to the prescription availability time of the previous batch of drugs; while abnormal drug purchases, in order to hoard a large number of drugs in a very short objective time, must immediately go to the next institution to continue to obtain drugs before the previous batch of drugs is theoretically far from being consumed, thus causing the virtual availability time of prescriptions to stack on the time axis. If the algorithm only adopts spatial features, it will miss this crucial difference in temporal evolution, leading to an abnormal amplification of the assessed distance for normal personnel movement. To address the false positive problem caused by a single spatial feature mapping, a cross-constraint mechanism based on the temporal dimension must be introduced. This mechanism requires extracting the overlap between the prescription cycle and the objective drug purchase timestamp. By calculating the degree of abnormal folding in prescription time, it can accurately distinguish between genuine medication needs and malicious cross-institutional drug hoarding, thereby providing reliable dynamic filtering gating for spatial features.

[0073] In summary, this invention first performs adjacent time-series correlation processing on the drug purchase timestamp sequence and the prescription availability days sequence in the medical insurance outpatient settlement feature data to obtain adjacent drug purchase time interval data and corresponding prescription availability duration data. Specifically, for any target patient, the drug purchase timestamp sequence and prescription availability days sequence corresponding to the target patient are extracted from the medical insurance outpatient settlement feature data, and the total number of drug purchase records corresponding to the target patient within a preset evaluation period is calculated. For any target drug purchase record in the drug purchase timestamp sequence corresponding to the target patient, excluding the first drug purchase record, the drug purchase timestamp corresponding to the target drug purchase record and the drug purchase timestamp corresponding to the previous drug purchase record are obtained. The difference between the drug purchase timestamp corresponding to the target drug purchase record and the drug purchase timestamp corresponding to the previous drug purchase record is used as the adjacent drug purchase time interval data corresponding to the target drug purchase record. The prescription availability days corresponding to the previous drug purchase record are extracted from the prescription availability days sequence corresponding to the target patient, and the prescription availability days corresponding to the previous drug purchase record are used as the prescription availability duration data corresponding to the target drug purchase record.

[0074] After obtaining the adjacent purchase time interval data and the corresponding prescription availability time data, the process continues by truncating and accumulating the adjacent purchase time interval data and the corresponding prescription availability time data to obtain cumulative prescription time overlap data. Specifically, for any target patient's purchase record other than the first purchase record, the adjacent purchase time interval data and prescription availability time data for the target purchase record are obtained. The difference between the prescription availability time data and the adjacent purchase time interval data for the target purchase record is used as the initial prescription time overlap assessment for the target purchase record. When the initial prescription time overlap assessment is greater than a constant 0, it is used as the prescription time overlap assessment for the target purchase record; when the initial prescription time overlap assessment is less than or equal to a constant 0, it is set to a constant 0. The prescription time overlap assessments for all purchase records for the target patient other than the first purchase record are summed, and the resulting sum is used as the cumulative prescription time overlap data for the target patient.

[0075] After obtaining the cumulative prescription time overlap data, the final step is to normalize and map the cumulative prescription time overlap data with the purchase time span data corresponding to the purchase timestamp sequence to obtain the spatiotemporal folding compression factor. Specifically, for any target patient, the purchase timestamp sequence corresponding to the target patient is extracted from the medical insurance outpatient settlement feature data. The purchase timestamps corresponding to the first and last purchase records in the purchase timestamp sequence are obtained. The difference between the purchase timestamps corresponding to the last and first purchase records is used as the initial purchase time span assessment for the target patient. When the initial purchase time span assessment is greater than or equal to a constant 1, the initial purchase time span assessment is used as the purchase time span data for the target patient; when the initial purchase time span assessment is less than a constant 1, the purchase time span data for the target patient is set to a constant 1. The cumulative prescription time overlap data for the target patient is obtained, and the result of adding the cumulative prescription time overlap data and the purchase time span data is used as the time folding normalization scale for the target patient. The cumulative prescription time overlap data is used as the numerator, the time folding normalization scale is used as the denominator, and the corresponding fraction is used as the spatiotemporal folding compression factor for the target patient.

[0076] In one embodiment, it is assumed that the total number of medication purchase records corresponding to the target patient within a preset evaluation period is [number missing]. ; Target patient The purchase timestamp corresponding to this medication purchase record is: ;No. The timestamp of the previous medication purchase record is: ;No. The prescription availability time data for the previous medication purchase record is as follows: The first medication purchase record in the medication purchase timestamp sequence corresponding to the target patient is... The purchase timestamp corresponding to the last purchase record in the purchase timestamp sequence for the target patient is... Then, the expression for calculating the spatiotemporal folding compression factor corresponding to the target patient is:

[0077]

[0078] in, This represents the spatiotemporal folding compression factor corresponding to the target patient; This indicates the total number of medication purchase records for the target patient within the preset assessment period; Indicates the target patient's number The purchase timestamp corresponding to each medication purchase record; Indicates the first The purchase timestamp corresponding to the previous purchase record of this medicine purchase record; Indicates the first Prescription availability duration data corresponding to the previous medication purchase record; This represents the purchase timestamp corresponding to the first purchase record in the purchase timestamp sequence for the target patient; This represents the purchase timestamp corresponding to the last purchase record in the purchase timestamp sequence for the target patient; This represents the maximum value function.

[0079] It should be noted that, considering the differences in the evolution of normal medication demand and malicious drug hoarding over time, this formula internally constructs an overlapping feature extraction structure formed by the difference between the number of days a prescription is available and the objective time interval. By incorporating a maximum value function, this structure implements a strict one-way conditional truncation logic. For a normal cross-institutional medication user group, the objective time interval between two medication purchases... It must be greater than or equal to the number of days the previous batch of medicines is available. At this point, the difference is negative, and the maximum value function directly outputs 0, completely eliminating the overlapping time of this normal medication use. For frequent violations occurring within a short period, with extremely short intervals between medication purchases, this item will accurately output the number of overlapping days of medication being prescribed before it was fully consumed. Through the outer layer summation operation, the model can extract the patient's cumulative prescription folding time throughout the entire assessment period. Simultaneously, this formula uses this cumulative folding time as the numerator, combining it with the objective duration spanning the beginning and end of the entire period. The summation is used as the normalized denominator. This projection structure based on the eigenvalue ratio can smoothly map the patient's temporal stacking to... Within the closed interval. When processing normal travel data that does not include advance medication stockpiling, the numerator is zero. A stable output of 0 completely eliminates any false penalty features that might be attached to the sample. However, when faced with extreme violations such as frequent roaming for medication prescriptions, the cumulative number of overlapping days far exceeds the objective time span, causing the output value of the spatiotemporal folding compression factor calculation formula to nonlinearly and rapidly approach the extreme value of 1. This time-series reconstruction process is entirely based on the inherent spatiotemporal evolution logic of medication records. By extracting the parameter-free features of the time folding rate, it provides a high-confidence correction benchmark for subsequent removal of spatial feature noise and reconstruction of distance metrics.

[0080] Thus, the spatiotemporal compression factor was obtained by performing time folding evaluation on the drug purchase timestamp sequence and the prescription usable days sequence in the medical insurance outpatient settlement feature data.

[0081] Step S4: Obtain the basic Mahalanobis distance by calculating the Mahalanobis distance of the global patient basic feature matrix, and obtain the reconstructed Mahalanobis distance by performing nonlinear joint correction on the basic Mahalanobis distance, spatial concealment camouflage factor and spatiotemporal folding compression factor.

[0082] After extracting the spatial concealment factor and the spatiotemporal folding compression factor, their joint features need to be fed back into the basic anomaly detection model to compensate for the shortcomings of Mahalanobis distance in processing spatiotemporal dimensionality reduction data. Existing Mahalanobis distance-based models rely solely on overall summary indicators to calculate outlier levels, failing to detect the amplified risks associated with roaming drug purchases. This embodiment employs a correction mechanism of penalty coefficient superposition, combining the correction factors obtained in steps S2 and S3 as multiplicative weights directly applied to the basic Mahalanobis distance calculated by the original model. However, considering that in real outpatient scenarios, normal patients may engage in a small amount of reasonable advance drug purchases due to holiday preparations or earlier appointment times, resulting in a slight non-zero value for their spatiotemporal folding compression factor. Directly imposing absolute penalties on this could easily lead to model oversensitivity and false alarms. Therefore, it is necessary to introduce an activation constraint mechanism based on nonlinear smoothing into the joint feature's action chain. This weakens the penalty weights caused by low-intensity overlap while retaining and amplifying the severe penalties caused by high-frequency extreme overlap, thereby providing a reasonable tolerance space for real minor drug preparation behaviors at the algorithm fusion level and ensuring the accuracy of the final reconstructed distance.

[0083] Specifically, based on the global patient baseline feature matrix, the mean values ​​of the total number of visits, total number of prescriptions, and total settlement amount for all patients within a preset assessment period are calculated to obtain a global mean feature vector. Then, based on the global patient baseline feature matrix, the variance and covariance among the baseline features of all patients are calculated to obtain a global covariance matrix and its corresponding inverse. For any target patient, the first type of summary feature corresponding to the target patient is extracted from the global patient baseline feature matrix, and a target patient baseline feature vector is constructed based on the total number of visits, total number of prescriptions, and total settlement amount. The difference between the target patient baseline feature vector and the global mean feature vector is used as the target patient's feature deviation vector. The transpose of the feature deviation vector, the global covariance inverse matrix, and the feature deviation vector are sequentially multiplied. The resulting calculation is used as the squared Mahalanobis distance assessment for the target patient, and the square root of the squared Mahalanobis distance assessment is used as the base Mahalanobis distance for the target patient.

[0084] The spatial concealment / camouflage factor and spatiotemporal folding compression factor corresponding to the target patient are obtained. The cube of the spatiotemporal folding compression factor is used as the temporal folding nonlinear evaluation corresponding to the target patient. The spatial concealment / camouflage factor and the temporal folding nonlinear evaluation are multiplied, and the resulting product is used as the spatiotemporal camouflage joint evaluation corresponding to the target patient. An exponential mapping with the natural constant as the base is applied to the spatiotemporal camouflage joint evaluation corresponding to the target patient, and the resulting mapping is used as the distance reconstruction amplification factor corresponding to the target patient. The basic Mahalanobis distance corresponding to the target patient is multiplied by the distance reconstruction amplification factor, and the resulting product is used as the reconstructed Mahalanobis distance corresponding to the target patient.

[0085] It should be noted that this embodiment achieves feature correction by constructing a distance reconstruction amplification factor. This design utilizes a cubic structure as a nonlinear mapping term for the spatiotemporal folding compression factor. Since the range of values ​​for the spatiotemporal folding compression factor is limited to... When normal patients experience a slight time overlap due to early follow-up visits, their spatiotemporal folding compression factor value is in a low range. After cubic operation, this value is nonlinearly reduced to near 0, causing a corresponding decrease in the spatiotemporal camouflage joint evaluation. Based on the natural exponential property, the spatiotemporal camouflage joint evaluation approaches 1 at this point, and the reconstructed Mahalanobis distance value approaches the basic Mahalanobis distance, thus preserving the basic judgment result of normal medication behavior during the feature reconstruction stage and providing a reasonable numerical tolerance range. Conversely, when processing data fragments of irregular drug purchases across institutions, the high frequency of drug acquisition in a short period leads to a large overlap in prescription time, causing the spatiotemporal folding compression factor value to rise to a high range. At this point, the reduction effect of cubic operation weakens, and combined with the increased spatial concealment camouflage factor due to the even distribution of drugs across institutions, the value of the spatiotemporal camouflage joint evaluation increases significantly. Under the influence of the natural constant, this structure can output nonlinearly amplified feature weights. This correction process overcomes the numerical constraints of linear calculations. When the basic Mahalanobis distance of a violation sample is lower than the system's judgment threshold due to the reduction of the single dose, the exponential mapping mechanism can effectively amplify its distance scale to above the anomaly judgment threshold, thereby achieving accurate extraction of the disguised features of dispersed drug purchases and effective identification of violation samples.

[0086] Thus far, the basic Mahalanobis distance has been obtained by calculating Mahalanobis distance from the global patient baseline feature matrix, and the reconstructed Mahalanobis distance has been obtained by nonlinearly combining the basic Mahalanobis distance, spatial concealment camouflage factor, and spatiotemporal folding compression factor.

[0087] Step S5: By comparing and analyzing the reconstructed Mahalanobis distance with the anomaly judgment threshold, the assessment results of medical insurance violations are obtained.

[0088] After obtaining the reconstructed Mahalanobis distance for each target patient, an anomaly assessment threshold is determined based on the three basic feature dimensions contained in the global patient basic feature matrix: total number of visits, total number of prescriptions, and total settlement amount. This threshold is determined according to the statistical property that the squared Mahalanobis distance follows a chi-square distribution. Preferably, the anomaly detection significance level is set to 0.01. The corresponding upper critical value is obtained according to the chi-square distribution with 3 degrees of freedom, and the square root of this critical value is taken as the anomaly assessment threshold.

[0089] For any target patient, the reconstructed Mahalanobis distance corresponding to the target patient is compared with an anomaly assessment threshold. When the reconstructed Mahalanobis distance is less than the anomaly assessment threshold, the target patient's outpatient medication purchase behavior is judged as normal; when the reconstructed Mahalanobis distance is greater than or equal to the anomaly assessment threshold, the target patient's outpatient medication purchase behavior is judged as abnormal, and the corresponding patient is marked as a high-risk abnormal sample. Based on the cross-medical institution medication purchase trajectory, prescription time overlap, and reconstructed Mahalanobis distance corresponding to the high-risk abnormal sample, violation warning analysis information is generated, thereby obtaining the assessment results of medical insurance violation behavior and providing data basis for subsequent medical insurance fund audit and abnormal behavior verification.

[0090] Thus, the assessment results of medical insurance violations were obtained by comparing and analyzing the reconstructed Mahalanobis distance with the anomaly judgment threshold.

[0091] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A data analysis and evaluation method for medical insurance violations, characterized in that, The method includes: Step S1: Collect and preprocess medical insurance outpatient settlement transaction data to obtain medical insurance outpatient settlement characteristic data; Step S2: Obtain the spatial concealment and camouflage factor by performing decentralized mapping processing on the cross-medical institution drug acquisition distribution characteristics in the medical insurance outpatient settlement characteristic data; Step S3: Obtain the spatiotemporal folding compression factor by performing time folding evaluation on the drug purchase timestamp sequence and the prescription usable days sequence in the medical insurance outpatient settlement feature data; Step S4: Obtain the basic Mahalanobis distance by calculating the Mahalanobis distance of the global patient basic feature matrix, and obtain the reconstructed Mahalanobis distance by performing nonlinear joint correction on the basic Mahalanobis distance, spatial concealment camouflage factor and spatiotemporal folding compression factor. Step S5: By comparing and analyzing the reconstructed Mahalanobis distance with the anomaly judgment threshold, the assessment results of medical insurance violations are obtained.

2. The data analysis and evaluation method for medical insurance violations according to claim 1, characterized in that, The process of collecting and preprocessing medical insurance outpatient settlement transaction data to obtain medical insurance outpatient settlement characteristic data includes: Obtain the outpatient chronic disease medication purchase settlement data recorded by the medical insurance settlement system within the target area during the preset evaluation period and perform data anonymization; for any outpatient chronic disease medication purchase settlement record, extract the patient's unique identification code, medical institution code, medication purchase timestamp, unified drug code, drug quantity, and settlement amount, and use the extracted data as medical insurance outpatient settlement data; Using the patient's unique identification code as the data aggregation dimension, abnormal records are removed and the time sequence is processed in the medical insurance outpatient settlement data. Outpatient chronic disease drug purchase settlement records with empty drug codes or abnormal drug prescription quantities are removed. Based on the drug instructions or the daily standard dosage of the corresponding drugs in the medical insurance catalog, the drug prescription quantities in each outpatient chronic disease drug purchase settlement record are converted into the corresponding prescription usage days. Outpatient chronic disease drug purchase settlement records for the same patient are arranged in chronological order of purchase timestamps. For any target patient, the total number of visits, total number of prescriptions, and total settlement amount are statistically obtained based on the outpatient chronic disease medication purchase settlement records within the preset assessment period. The total number of visits, total number of prescriptions, and total settlement amount are used as the first type of summary feature of the target patient. For the target drug, the total number of different medical institutions visited by the target patient, the cumulative total quantity of the target drug, and the maximum quantity of the target drug prescribed by a single medical institution are statistically obtained. At the same time, the medication purchase timestamp sequence and the prescription validity period sequence corresponding to the target patient are obtained. The total number of different medical institutions, the cumulative total quantity of the target drug, the maximum quantity of the target drug prescribed by a single medical institution, the medication purchase timestamp sequence, and the prescription validity period sequence are used as the second type of spatiotemporal trajectory feature of the target patient. The first type of aggregated features of all patients within the preset assessment period are combined according to the patient dimension to obtain the global patient basic feature matrix. The global patient basic feature matrix and the second type of spatiotemporal trajectory features corresponding to all patients are used as medical insurance outpatient settlement feature data.

3. The data analysis and evaluation method for medical insurance violations according to claim 1, characterized in that, The process of obtaining spatial concealment and camouflage factors by decentralizing the distribution characteristics of drug acquisition across medical institutions in the medical insurance outpatient settlement feature data includes: By performing cross-institutional distribution feature extraction on the total number of medical institutions visited and the distribution of drug acquisition quantity in the medical insurance outpatient settlement feature data, institutional divergence constraint data and drug decentralization depth data are obtained. By cross-coupling and exponential mapping of institutional divergence constraint data and decentralized drug depth data, spatial concealment and camouflage factors are obtained.

4. The data analysis and evaluation method for medical insurance violations according to claim 3, characterized in that, The process involves extracting cross-institutional distribution features from the total number of medical institutions visited and the distribution of drug purchases in the medical insurance outpatient settlement feature data to obtain institutional divergence constraint data and drug decentralization depth data, including: For any target patient, extract the total number of medical institutions visited by the target patient, the total number of target drugs, and the maximum number of target drugs prescribed by a single medical institution from the medical insurance outpatient settlement feature data. The difference between the total number of medical institutions visited by the target patient and the constant 1 is used as the evaluation of the number of institutions dispersed for the target patient. The evaluation of the number of institutions dispersed is used as the numerator, the total number of medical institutions visited is used as the denominator, and the resulting fraction is used as the institutional divergence constraint data for the target patient. The difference between the total cumulative quantity of target drugs for target patients and the maximum quantity of target drugs prescribed by a single medical institution is used as the drug quantity assessment for non-primary medical institutions for target patients. The drug quantity assessment for non-primary medical institutions is used as the numerator, and the total cumulative quantity of target drugs is used as the denominator. The resulting fraction is used as the decentralized deep data of drugs for target patients.

5. The data analysis and evaluation method for medical insurance violations according to claim 3, characterized in that, The process of obtaining spatial concealment and camouflage factors by cross-coupling and exponential mapping between institutional divergence constraint data and decentralized drug depth data includes: For any target patient, obtain the institutional divergence constraint data and the drug decentralized depth data corresponding to the target patient; Multiply the institutional divergence constraint data corresponding to the target patient with the decentralized depth data of the drug, and use the resulting product as the cross-institutional decentralized cross-evaluation corresponding to the target patient. An exponential mapping with a natural constant as the base is performed on the cross-institutional decentralized cross-assessment corresponding to the target patient, and the corresponding mapping result is used as the spatial concealment camouflage factor corresponding to the target patient.

6. The data analysis and evaluation method for medical insurance violations according to claim 1, characterized in that, The process involves performing time-folding evaluation on the drug purchase timestamp sequence and the prescription's allowed usable days sequence in the medical insurance outpatient settlement characteristic data to obtain a spatiotemporal folding compression factor, including: By performing adjacent time-series correlation processing on the drug purchase timestamp sequence and the prescription usable days sequence in the medical insurance outpatient settlement feature data, adjacent drug purchase time interval data and corresponding prescription usable duration data are obtained. By performing time overlap truncation and accumulation processing on adjacent drug purchase time interval data and corresponding prescription availability time data, cumulative prescription time overlap data is obtained. By normalizing and mapping the cumulative prescription time overlap data with the drug purchase time span data corresponding to the drug purchase timestamp sequence, the spatiotemporal folding compression factor is obtained.

7. The data analysis and evaluation method for medical insurance violations according to claim 6, characterized in that, The process involves performing adjacent time-series correlation processing on the drug purchase timestamp sequence and the prescription validity period sequence in the medical insurance outpatient settlement feature data to obtain adjacent drug purchase time interval data and corresponding prescription validity period data, including: For any target patient, extract the corresponding drug purchase timestamp sequence and prescription validity period sequence from the medical insurance outpatient settlement feature data, and count the total number of drug purchase records corresponding to the target patient within the preset assessment period; For any target drug purchase record other than the first drug purchase record in the drug purchase timestamp sequence corresponding to the target patient, obtain the drug purchase timestamp corresponding to the target drug purchase record and the drug purchase timestamp corresponding to the previous drug purchase record. Use the difference between the drug purchase timestamp corresponding to the target drug purchase record and the drug purchase timestamp corresponding to the previous drug purchase record as the adjacent drug purchase time interval data corresponding to the target drug purchase record. Extract the number of days the prescription is available from the sequence of days the prescription is available for the target patient, and use the number of days the prescription is available for the previous purchase record as the prescription availability duration data for the target purchase record.

8. The data analysis and evaluation method for medical insurance violations according to claim 6, characterized in that, The process of truncating and accumulating adjacent drug purchase time interval data and corresponding prescription availability time data to obtain cumulative prescription time overlap data includes: For any target patient and any target medication purchase record other than the first medication purchase record, obtain the adjacent medication purchase time interval data and prescription availability duration data corresponding to the target medication purchase record; The difference between the available prescription duration data corresponding to the target drug purchase record and the adjacent drug purchase time interval data is used as the initial prescription time overlap assessment for the target drug purchase record; when the initial prescription time overlap assessment is greater than the constant 0, the initial prescription time overlap assessment is used as the prescription time overlap assessment for the target drug purchase record; when the initial prescription time overlap assessment is less than or equal to the constant 0, the prescription time overlap assessment for the target drug purchase record is set to the constant 0. The prescription time overlap assessments for all medication purchase records (excluding the first purchase record) corresponding to the target patient are summed up, and the resulting sum is used as the cumulative prescription time overlap data for the target patient.

9. The data analysis and evaluation method for medical insurance violations according to claim 6, characterized in that, The process of obtaining a spatiotemporal folding compression factor by normalizing and mapping the cumulative prescription time overlap data with the purchase time span data corresponding to the purchase timestamp sequence includes: For any target patient, extract the drug purchase timestamp sequence corresponding to the target patient from the medical insurance outpatient settlement feature data, and obtain the drug purchase timestamp corresponding to the first drug purchase record and the drug purchase timestamp corresponding to the last drug purchase record in the drug purchase timestamp sequence. The difference between the timestamp of the last medication purchase record and the timestamp of the first medication purchase record is used as the initial medication purchase time span assessment for the target patient. When the initial medication purchase time span assessment is greater than or equal to a constant 1, the initial medication purchase time span assessment is used as the medication purchase time span data for the target patient. When the initial medication purchase time span assessment is less than a constant 1, the medication purchase time span data for the target patient is set to a constant 1. Obtain the cumulative prescription time overlap data corresponding to the target patient, and use the calculation result of adding the cumulative prescription time overlap data and the drug purchase time span data as the time folding normalization scale corresponding to the target patient; use the cumulative prescription time overlap data as the numerator, the time folding normalization scale as the denominator, and use the corresponding fraction as the spatiotemporal folding compression factor corresponding to the target patient.

10. The data analysis and evaluation method for medical insurance violations according to claim 1, characterized in that, The process involves calculating the basic Mahalanobis distance from the global patient baseline feature matrix, and then obtaining the reconstructed Mahalanobis distance through nonlinear joint correction of the basic Mahalanobis distance, spatial concealment factor, and spatiotemporal folding compression factor. This includes: Based on the global patient basic feature matrix, the mean of the total number of visits, total number of prescriptions and total settlement amount for all patients within the preset assessment period is calculated to obtain the global mean feature vector. Based on the global patient basic feature matrix, the variance and covariance between the basic features of all patients are calculated to obtain the global covariance matrix and the corresponding global covariance inverse matrix. For any target patient, extract the first type of summary features corresponding to the target patient from the global patient basic feature matrix, and construct the target patient basic feature vector based on the total number of visits, total number of prescriptions, and total settlement amount corresponding to the target patient; take the difference between the target patient basic feature vector and the global mean feature vector as the target patient's feature deviation vector, perform matrix multiplication on the transpose of the feature deviation vector, the global covariance inverse matrix, and the feature deviation vector in sequence, take the corresponding calculation result as the squared Mahalanobis distance evaluation corresponding to the target patient, and take the square root of the squared Mahalanobis distance evaluation as the basic Mahalanobis distance corresponding to the target patient; Obtain the spatial concealment and camouflage factor and the spatiotemporal folding compression factor corresponding to the target patient. Use the cube of the spatiotemporal folding compression factor as the time folding nonlinear assessment corresponding to the target patient. Multiply the spatial concealment and camouflage factor and the time folding nonlinear assessment. Use the corresponding product as the spatiotemporal camouflage joint assessment corresponding to the target patient. The spatiotemporal camouflage joint assessment corresponding to the target patient is subjected to an exponential mapping with the natural constant as the base, and the corresponding mapping result is used as the distance reconstruction amplification factor corresponding to the target patient; the basic Mahalanobis distance corresponding to the target patient is multiplied by the distance reconstruction amplification factor, and the corresponding product is used as the reconstructed Mahalanobis distance corresponding to the target patient.

Images