Method for collaborative optimization based on medical data resource management and intelligent analysis

Abstract

The application relates to the technical field of medical data processing, and particularly discloses a medical data resource management and intelligent analysis collaborative optimization method, which is used for collecting multi-source heterogeneous blood pressure data streams in real time and obtaining a reference distribution; for each batch of paired blood pressure fields, the likelihood matching degrees of original and exchanged combinations under the reference distribution are calculated, and a likelihood ratio set is constructed; the field reversal abnormality is determined according to the median deviation degree and direction consistency of the likelihood ratio set, and the probability migration distance of the data distribution after correction and the reference distribution is measured; when the abnormality is established and the distance is lower than the upper limit of tolerance, field exchange is performed, the data stream is reorganized and corrected based on implicit order identification and prime number mapping, and a temporary mapping hypothesis is established; subsequent batches are assigned with decreasing weights, and the temporary hypothesis is upgraded to a permanent rule or transferred to manual verification through a double threshold of weighted effective counting and absolute quantity; the application realizes online repair of the field reversal abnormality and continuous verification of the hypothesis, and avoids data flow interruption.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, specifically to a collaborative optimization method based on medical data resource governance and intelligent analysis. Background Technology

[0002] In the construction of regional medical information systems, the platform needs to access heterogeneous data sources from hospitals, community centers, and physical examination institutions of different levels. Among these, blood pressure measurement data is one of the most common physiological indicators. When outputting blood pressure records, the information systems of different institutions (such as electronic medical record systems and laboratory information systems) often organize the numerical fields according to their own internal protocols. For example, some systems use systolic blood pressure as the first field and diastolic blood pressure as the second field, while other systems do the opposite. In some older systems, the semantics of the output fields are even unknown.

[0003] When the intelligent analysis model detects distribution anomalies in multi-source heterogeneous blood pressure data, the existing governance system can only perform coarse-grained operations such as rejecting or forcibly supplementing data. It cannot identify the root cause of the anomaly as field inversion at the semantic level, and therefore cannot automatically complete field exchange repair and output corrected data streams, resulting in the accidental interception and interruption of critical business data. At the same time, the existing methods lack a mechanism for weighted decremental verification and dual threshold upgrade of temporary repair hypotheses, making it difficult to strike a balance between avoiding the solidification of erroneous rules and ensuring the continuity of analysis. Summary of the Invention

[0004] The purpose of this invention is to provide a collaborative optimization method based on medical data resource governance and intelligent analysis to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions: The collaborative optimization method based on medical data resource governance and intelligent analysis includes the following steps: S1: Real-time acquisition of continuous data streams from multiple heterogeneous medical data sources, each of which includes a pair of blood pressure value fields, while also acquiring high-quality reference data distributions provided by a verified interface; S2: For each pair of blood pressure value fields in the current batch, calculate the likelihood matching degree of the original field combination and the exchanged field combination under the reference data distribution, and construct a set of likelihood ratios based on the matching degree difference of all samples. S3: Based on the median deviation of the likelihood ratio set and the consistency of the deviation direction of each sample, determine whether there is a field inversion anomaly in the current batch. At the same time, use probability migration distance to calculate the difference between the corrected data distribution and the reference data distribution, and determine whether the corresponding difference is lower than the tolerance limit. S4: When the median deviates beyond the threshold and the probability migration distance is below the tolerance limit, the field swap operation is automatically executed to generate corrected data, the corrected data is output to the subsequent analysis process, and a temporary mapping hypothesis record is established to record the corresponding swap relationship; S5: Continuously monitor the continued validity of the temporary mapping hypothesis and the probability migration distance of the corrected data in each batch in subsequent batches. When the cumulative number of valid batches exceeds the preset threshold and the migration distance is always lower than the tolerance limit, upgrade the temporary mapping hypothesis to a permanently effective mapping rule; otherwise, trigger a re-judgment or switch to the manual verification path.

[0006] As a further aspect of the present invention: S2 specifically includes: Discretize the continuous probability density of the reference data distribution into multiple equally probable regions; For each pair of blood pressure value fields in the current batch, locate the density level of the original field combination and the exchanged field combination falling into the equally probable region, and calculate the density ratio of the two combinations in the same level. After taking the natural logarithm of the density ratios of all samples in the corresponding batch, sort them by numerical value and select a preset number of logarithmic values ​​in the middle position as the likelihood ratio set.

[0007] As a further aspect of the present invention: the calculation of the density ratio of the two combinations at the same level specifically includes: The two-dimensional space of the reference data distribution is divided at equal intervals according to the cumulative probability of each dimension, forming multiple rectangular cells with equal probability areas, and each cell is pre-assigned a hierarchical number; Determine the rectangular cells where the original field combination and the swapped field combination fall respectively, and read the corresponding two hierarchical numbers; The ratio of the smaller value to the larger value of the two hierarchical indices is taken as the density ratio of the corresponding samples.

[0008] As a further aspect of the present invention: S3 specifically includes: The relative deviation index is obtained by calculating the ratio of the median of the likelihood ratio set to the absolute median difference of the likelihood ratio set. The ratio of the difference between the number of samples with a value greater than 0 and the number of samples with a value less than 0 in the statistical likelihood ratio set is taken as the directional consistency coefficient. When the relative deviation index exceeds the preset threshold and the directional consistency coefficient is higher than the effective threshold, it is determined that there is a field inversion anomaly in the current batch.

[0009] As a further aspect of the present invention: S4 specifically includes: Read the original order of each pair of blood pressure value fields in the current batch, swap the positions of the first and second fields, and generate the swapped field pair; The swapped field pairs are reassembled into a corrected data stream in the original record order, and the corrected data stream is passed to the subsequent analysis stage; Using the interface identifier of the current data source and the current time window as a composite primary key, record a temporary mapping hypothesis that includes the exchange operation type and the effective monitoring period length.

[0010] As a further aspect of the present invention: the reassembly of the swapped field pairs into a corrected data stream according to the original record order specifically includes: Extract the implicit sequence identifier attached to each original record in the current batch. The implicit sequence identifier is the numerical value of the arrival order of the record in the data stream. The modified field obtained by exchanging fields in each record is bound to the implicit sequence identifier of the records to form an ordered entry; The data is rearranged in ascending order based on the numerical values ​​of the implicit sequence identifiers in all ordered entries. The rearranged correction field pairs are then concatenated sequentially to obtain the corrected data stream.

[0011] As a further aspect of the present invention: the step of binding the modified field obtained by exchanging fields of each record with the implicit sequence identifier of the records to form an ordered entry specifically includes: A prime number mapping table is pre-defined, which corresponds one-to-one with the value space of the implicit sequence identifier. The corresponding prime number is found based on the implicit sequence identifier of the current record. The first value in the correction field pair is multiplied by the found prime number, and the last six digits of the product are taken as the first binding value. The second value in the correction field pair is multiplied by the same prime number, and the last six digits of the product are taken as the second binding value. The implicit sequence identifier, the first binding value, and the second binding value are concatenated into an ordered entry in chronological order.

[0012] As a further aspect of the present invention: S5 specifically includes: Assign a weight value to each subsequent batch that decreases as the batch number increases, and normalize the sum of all weight values ​​to a unit of 1; Determine whether the probability migration distance of each subsequent batch is lower than the tolerance limit. Record the batches with the result lower as valid batches and add the weight value corresponding to the valid batches to the weighted valid count. When the weighted valid count exceeds the preset cumulative threshold, and the absolute number of valid batches also reaches the minimum number threshold, the cumulative number of valid batches is determined to meet the upgrade conditions.

[0013] The beneficial effects of this invention are: (1) Upon detecting a field inversion anomaly, this invention does not directly reject or return the abnormal data. Instead, it calculates the hierarchical ratio and probability migration distance of the original and swapped field combinations under the reference distribution. When the anomaly is confirmed, it automatically performs a field swap and outputs the corrected data stream to the subsequent analysis stage. Simultaneously, it reassembles the data stream using implicit sequence identifiers and a prime number mapping multiplication and remainder method, ensuring the restoration of record order under out-of-order transmission. This avoids the problem of data interruption on the regional health platform caused by abrupt rejection in existing technologies, enabling abnormal blood pressure data from older systems to be corrected online without interrupting business analysis, thus ensuring the continuous operation of data governance and analysis.

[0014] (2) After establishing a temporary mapping hypothesis, this invention assigns weight values ​​that decrease with the batch number to subsequent batches, and calculates the weighted effective count and the absolute number of effective batches. Only when both exceed a preset threshold is the temporary hypothesis upgraded to a permanently effective mapping rule. This mechanism gives higher weight to the verification results of recent batches, while the contribution of earlier batches decreases, avoiding the solidification of erroneous rules due to short-term accidental compliance. When the upgrade conditions are not met, the system either discards the hypothesis or transfers it to manual verification based on the number of effective batches, thereby achieving a balance between automation and security, reducing the frequency of manual review, and preventing erroneous mapping rules from taking effect for a long time. Attached Figure Description

[0015] The invention will now be further described with reference to the accompanying drawings.

[0016] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 As shown, this invention is a collaborative optimization method based on medical data resource governance and intelligent analysis, comprising the following steps: S1: Real-time acquisition of continuous data streams from multiple heterogeneous medical data sources, each of which includes a pair of blood pressure value fields, while also acquiring high-quality reference data distributions provided by a verified interface; S2: For each pair of blood pressure value fields in the current batch, calculate the likelihood matching degree of the original field combination and the exchanged field combination under the reference data distribution, and construct a set of likelihood ratios based on the matching degree difference of all samples. S3: Based on the median deviation of the likelihood ratio set and the consistency of the deviation direction of each sample, determine whether there is a field inversion anomaly in the current batch. At the same time, use probability migration distance to calculate the difference between the corrected data distribution and the reference data distribution, and determine whether the corresponding difference is lower than the tolerance limit. S4: When the median deviates beyond the threshold and the probability migration distance is below the tolerance limit, the field swap operation is automatically executed to generate corrected data, the corrected data is output to the subsequent analysis process, and a temporary mapping hypothesis record is established to record the corresponding swap relationship; S5: Continuously monitor the continued validity of the temporary mapping hypothesis and the probability migration distance of the corrected data in each batch in subsequent batches. When the cumulative number of valid batches exceeds the preset threshold and the migration distance is always lower than the tolerance limit, upgrade the temporary mapping hypothesis to a permanently effective mapping rule; otherwise, trigger a re-judgment or switch to the manual verification path.

[0019] In S1, continuous data streams from multiple heterogeneous medical data sources are acquired in real time. Each continuous data stream includes paired blood pressure value fields. Simultaneously, high-quality reference data distributions provided by a verified interface are obtained, specifically including: Real-time acquisition of continuous data streams from multiple heterogeneous medical data sources is achieved through the following method: A data acquisition gateway is deployed at the data exit point of hospitals, community centers, or physical examination institutions. This gateway captures newly generated blood pressure measurement records from internal electronic medical record systems or laboratory information systems in real time through a database log reading interface. Each record contains at least two numerical fields, a patient identifier, and an arrival timestamp. The data acquisition gateway encapsulates the captured records into lightweight messages in arrival order and continuously pushes them to the unified access port of the regional health platform through a message queue middleware, forming a continuous data stream. Each pair of blood pressure numerical fields maintains its original, semantically unknown, arrangement order within the message body. Simultaneously, the platform obtains high-quality reference data distribution from a verified interface: this verified interface connects to the information system of a tertiary-level hospital, whose blood pressure field mapping relationship has been manually verified line by line, confirming that the first numerical field is always systolic pressure and the second numerical field is always diastolic pressure. The platform continuously collects 10,000 verified blood pressure records from this interface, constructs a two-dimensional continuous probability density function using the Gaussian kernel density estimation method, and stores it as the reference data distribution for likelihood calculation and distribution comparison in subsequent steps.

[0020] In S2, for each pair of blood pressure value fields in the current batch, the likelihood matching degree of the original field combination and the exchanged field combination under the reference data distribution is calculated, and a likelihood ratio set is constructed based on the matching degree difference of all samples, specifically including: The continuous probability density of the reference data distribution is discretized into multiple equally probable regions. Specifically, based on the two-dimensional cumulative distribution function corresponding to the reference data distribution, the cumulative probability is divided into 10 equally spaced intervals from 0 to 1 for both systolic and diastolic blood pressure dimensions, with each interval having a cumulative probability width of 0.1. Thus, the interval divisions in the two dimensions are orthogonal, forming 100 rectangular cells (10 x 10), each covering an equal probability area of ​​0.01. For each rectangular cell, the probability density value at its geometric center is calculated. Then, all 100 cells are sorted from highest to lowest probability density value, with the cell with the highest density assigned a hierarchy number of 1, the second highest as 2, and so on, down to the cell with the lowest density assigned a hierarchy number of 100.

[0021] For each pair of blood pressure values ​​in the current batch, locate the positions of the original field combination and the swapped field combination within the rectangular cells, and read the corresponding hierarchical sequence number. The specific process is as follows: For the original field combination in each record, i.e., the first and second values, determine the interval number of the first value in the cumulative probability of systolic blood pressure and the interval number of the second value in the cumulative probability of diastolic blood pressure. These two interval numbers uniquely identify a rectangular cell, and read the pre-assigned hierarchical sequence number of this cell, denoted as the first hierarchical sequence number. Then, for the swapped field combination of the record, i.e., the second value as the first value and the first value as the second value, similarly determine their interval numbers in both dimensions, find the corresponding rectangular cell, and read its hierarchical sequence number, denoted as the second hierarchical sequence number.

[0022] After obtaining the first-level and second-level indices of the same record, the density ratio of that record is calculated. This density ratio is defined as the quotient obtained by dividing the smaller of the first-level and second-level indices by the larger one. Since both level indices are positive integers between 1 and 100, the quotient is a real number between 0 and 1. If the first-level and second-level indices are equal, the density ratio is equal to 1; if they are not equal, the density ratio is less than 1. This density ratio reflects the likelihood of the original field combination relative to the exchanged field combination under the reference data distribution. The closer the ratio is to 1, the closer the probability density levels of the two combinations are under that distribution.

[0023] After calculating the density ratio for each record in the current batch, the natural logarithm of the density ratios for all samples in the batch is taken. In the natural logarithm calculation, the logarithm value is 0 when the density ratio is equal to 1, and negative when the density ratio is less than 1. All natural logarithm values ​​are sorted in ascending order to obtain an increasing sequence. Then, the smallest value in the top 25% and the largest value in the bottom 25% are removed from this sequence, retaining only the middle 50% of the logarithm values. These retained logarithm values ​​form a set called the likelihood ratio set. This preset size is half the total number of samples in the current batch. If the total number of samples in the current batch is less than 100, all logarithm values ​​are retained without removal.

[0024] In the above process, parameters such as 10 equal intervals, 100 cells, hierarchical numbers from 1 to 100, and retaining the middle 50% are all preset fixed values ​​that can be adjusted according to the actual data scale. For example, if the sample size of the reference data distribution exceeds 100,000, the number of equal intervals for each dimension can be increased to 20, forming 400 cells, with a corresponding hierarchical number range of 1 to 400. All numerical calculations retain six decimal places and are stored with floating-point precision for use in subsequent steps.

[0025] By employing discretization and hierarchical ratio calculation methods, the likelihood matching problem between the original field combination and the exchanged field combination under continuous probability density is transformed into a ratio problem of discrete cell hierarchical indices. This eliminates the need for complex integral calculations, significantly reducing computational resource consumption while maintaining the ability to distinguish distribution similarities. The resulting set of likelihood ratios provides a stable and robust statistical basis for subsequent anomaly detection.

[0026] In S3, based on the median deviation of the likelihood ratio set and the consistency of the deviation direction of each sample, it is determined whether there is a field inversion anomaly in the current batch. Simultaneously, the probability migration distance is used to calculate the difference between the corrected data distribution and the reference data distribution, and it is determined whether the corresponding difference is below the tolerance limit. Specifically, this includes: The deviation of the median is calculated based on the likelihood ratio set. Specifically, all values ​​in the likelihood ratio set are arranged in ascending order, and the value in the middle position is taken as the median. If the number of values ​​in the likelihood ratio set is even, the arithmetic mean of the two middle values ​​is taken as the median. Then, the absolute value of the difference between each likelihood ratio value and the median is calculated, resulting in a set of absolute differences. These absolute differences are arranged in ascending order, and the value in the middle position is taken as the absolute median difference. Finally, the absolute value of the median is divided by the absolute median difference, and the quotient is defined as the relative deviation index. If the absolute median difference is zero, the relative deviation index is directly set to a large value greater than a threshold, such as 1000, to indicate the existence of a completely consistent deviation.

[0027] After obtaining the relative deviation index, the consistency of deviation direction for each sample in the likelihood ratio set is simultaneously calculated. The specific process is as follows: Iterate through each value in the likelihood ratio set, counting the number of values ​​greater than zero (referred to as the positive quantity) and the number of values ​​less than zero (referred to as the negative quantity); values ​​equal to zero are not counted. Then, calculate the absolute value of the difference between the positive and negative quantities, divide this difference by the sum of the positive and negative quantities, and define the resulting ratio as the direction consistency coefficient. This coefficient ranges from 0 to 1. The closer the coefficient is to 1, the more likely most values ​​are either all positive or all negative, indicating a high degree of consistency in the deviation direction; the closer the coefficient is to 0, the more likely the positive and negative quantities are roughly equal, indicating a more varied and chaotic deviation direction.

[0028] Two thresholds are set: the relative deviation index is preset to 3.0, and the directional consistency coefficient is set to 0.7. When both the relative deviation index and the directional consistency coefficient are greater than 3.0 and greater than 0.7, the current batch is determined to have a field inversion anomaly. If either of these two conditions is not met, the current batch is determined not to have a field inversion anomaly, and no subsequent swapping operation is triggered; the original field combination is directly used for downstream analysis.

[0029] While determining the field inversion anomaly, it is also necessary to calculate the probability migration distance between the corrected data distribution and the reference data distribution. The specific method is as follows: Perform a swap operation on the original field combinations of each record in the current batch to obtain corrected field pairs. Use these corrected field pairs as sample points to construct a temporary two-dimensional empirical distribution. Then, calculate the bulldozer distance between this empirical distribution and the reference data distribution. This distance is defined as the minimum average displacement required to migrate the probability mass of the empirical distribution to the reference data distribution. In this embodiment, a simplified calculation method is used: calculate the envelope area between the cumulative distribution function of the corrected sample and the cumulative distribution function of the reference distribution in both the systolic and diastolic blood pressure dimensions. Add the areas in the two dimensions and divide by 2 to obtain an approximate value of the probability migration distance.

[0030] The preset tolerance limit is 0.15, which is determined by multiplying the maximum probability migration distance obtained after performing the same calculation on known healthy hospital data by 1.5. If the calculated probability migration distance is less than the tolerance limit of 0.15, the difference is considered to be below the tolerance limit; otherwise, it is considered to be not less than the tolerance limit. Only when both conditions are met—the field inversion anomaly determination and the probability migration distance being less than the tolerance limit—will the subsequent correction and mapping record steps proceed; if either condition is not met, the swap operation is skipped, the original data remains unchanged, and the anomaly in this batch is marked as awaiting manual review.

[0031] In the above calculations, the relative deviation index threshold of 3.0, the directional consistency coefficient threshold of 0.7, and the upper limit of probability migration distance tolerance of 0.15 are all configurable parameters that have been verified through multiple tests. In practical applications, they can be fine-tuned according to the overall quality of the access data source. All numerical calculations use double-precision floating-point numbers, retaining six decimal places to ensure the accuracy and stability of the judgment.

[0032] In S4, when the median deviates beyond a threshold and the probability migration distance is below the tolerance limit, a field swap operation is automatically performed to generate corrected data. This corrected data is then output to subsequent analysis steps, and a temporary mapping hypothesis is established to record the corresponding swap relationship. Specifically, this includes: When the relative deviation index exceeds 3.0 and the probability migration distance is less than 0.15, a field inversion anomaly is determined, and a field swap operation is automatically executed. The specific process is as follows: The original order of each pair of blood pressure value fields in the current batch is read, i.e., the first and second arriving values ​​in each record. The values ​​of the first and second fields are swapped, generating a new field pair where the original second value becomes the new first value, and the original first value becomes the new second value. This swap operation does not change the stored content of the original record; it only generates a copy of the swapped field pair in the data stream processing memory.

[0033] To reassemble the swapped field pairs into a corrected data stream in the original record order, it is first necessary to extract the implicit sequence identifier attached to each original record in the current batch. This implicit sequence identifier is generated by the data acquisition gateway when encapsulating each blood pressure measurement record. Specifically, it is an integer sequence number that increments sequentially from 1, with each sequence number uniquely corresponding to one record. This sequence number is pushed to the access port along with the message. When the data stream becomes out of order due to network transmission or message queue processing, this implicit sequence identifier still retains the original order information of the records when they were acquired. For each record in the current batch, its attached implicit sequence identifier is read and recorded as a sequence number.

[0034] Each record's modified field, obtained through field swapping, is bound to its implicit sequence identifier to form an ordered entry. The binding process uses prime number mapping and multiplication with remainder. First, a prime number mapping table is pre-built, mapping each possible implicit sequence identifier value to a unique prime number. The prime number mapping table is constructed as follows: starting from 2, the first prime number 2 corresponds to sequence number 1; the second prime number 3 corresponds to sequence number 2; the third prime number 5 corresponds to sequence number 3; the fourth prime number 7 corresponds to sequence number 4; the fifth prime number 11 corresponds to sequence number 5; and so on, mapping in ascending order of prime numbers. Since the number of samples in the current batch is usually no more than 1000, pre-storing the first 1000 prime numbers is sufficient. Based on the implicit sequence identifier value of the current record, the corresponding prime number is found in the mapping table and denoted as the mapped prime number.

[0035] Perform a multiplication and modulo operation on the first and second values ​​in the correction field pair respectively. Let the first value in the correction field pair be the first value, and the mapped prime number be the prime number parameter. The formula for calculating the first bound value is as follows: ; in, Indicates the first bound value. This indicates the first value in the correction field pair (in millimeters of mercury, taking the integer part). Represents a mapping prime number. This represents the remainder operation, i.e., calculation. The remainder after multiplying the product of P and P by 1,000,000. The remainder is an integer between 0 and 999,999, padded with zeros if less than six digits. Similarly, let the second value in the correction field pair be the second value. The formula for calculating the second bound value is as follows: ; Here, B2 represents the second binding value, V2 represents the second value in the correction field pair, and P is the same mapping prime number. This operation multiplies the original physiological value by the prime number and truncates it to the last six digits, eliminating the direct comparability between different values ​​while retaining the ability to distinguish between different records.

[0036] After completing the above calculations, the implicit sequence identifier, first binding value, and second binding value of the current record are concatenated into an ordered entry in chronological order. The specific concatenation format is as follows: first, write the value of the implicit sequence identifier; then, write a separator (e.g., an underscore character); next, write the six-digit number of the first binding value (padding with zeros if less than six digits); then write the separator again; and finally, write the six-digit number of the second binding value. For example, if the implicit sequence identifier is 3, the first binding value is 000123, and the second binding value is 456789, then the ordered entry is represented as "3_000123_456789". This ordered entry contains the original order information and the corrected numerical binding result.

[0037] After all ordered entries have been generated, they are rearranged in ascending order based on the implicit order identifier values ​​in each entry. Specifically, all ordered entries are rearranged according to their implicit order identifiers from smallest to largest, restoring the records that were previously misaligned due to network out-of-order processing to their original order at the time of acquisition. Then, the first and second binding values ​​of each record in the rearranged order are extracted sequentially, and these two values ​​are concatenated in their original order to form a continuous corrected data stream. Each pair of values ​​in this corrected data stream corresponds to a blood pressure record that has undergone field swapping and order restoration, and is subsequently passed to subsequent analysis stages, such as hypertension prevalence statistics or risk stratification assessment.

[0038] Finally, a temporary mapping hypothesis is recorded using the interface identifier of the current data source and the current time window as a composite primary key. The interface identifier is a unique code for the secondary hospital or medical examination institution, such as "H0023". The current time window is defined as the start and end timestamps of the batch. The start timestamp is the arrival time of the first record in the batch, and the end timestamp is the arrival time of the last record, in millisecond-level Unix timestamp format. The interface identifier and time window are concatenated as a composite primary key. The temporary mapping hypothesis record includes: the swap operation type, with a value of "field inversion swap"; and the effective monitoring period length, set to the next 10 batches. This record is stored in local persistent storage for continuous monitoring in step S5.

[0039] In S5, the continued validity of the temporary mapping hypothesis and the probability migration distance of the corrected data in each batch are continuously monitored across multiple subsequent batches. When the cumulative number of valid batches exceeds a preset threshold and the migration distance remains below the tolerance limit, the temporary mapping hypothesis is upgraded to a permanently effective mapping rule; otherwise, a re-evaluation or transfer to manual verification is triggered, specifically including: Each subsequent batch is assigned a weight value that decreases as the batch number increases. Specifically, starting from the batch after the temporary mapping assumption is established, subsequent batches are numbered sequentially by time as the 1st subsequent batch, the 2nd subsequent batch, and so on, up to the 10th subsequent batch. The weight value of the 1st subsequent batch is set to 10 divided by 55, the weight value of the 2nd subsequent batch is set to 9 divided by 55, the weight value of the 3rd subsequent batch is set to 8 divided by 55, and so on, with the weight value of the 10th subsequent batch set to 1 divided by 55. The denominator 55 is the sum of the values ​​from 1 to 10. In this way, the sum of the weight values ​​of all 10 batches is exactly equal to 1, i.e., normalized to a unit of 1. If, during actual monitoring, a data source stops sending data, resulting in fewer than 10 monitored batches, then only the batches that actually arrive are recalculated according to the above rules, so that the sum of the weights of the remaining batches is still 1.

[0040] Determine if the probability migration distance of each subsequent batch is lower than the tolerance limit of 0.15. For each subsequent batch, calculate the probability migration distance between the corrected data of that batch and the reference data distribution according to the method described in step S3, and compare this distance with 0.15. If the probability migration distance of a batch is less than 0.15, the batch is recorded as a valid batch; if it is greater than or equal to 0.15, it is recorded as an invalid batch. For each valid batch, its corresponding weight value is accumulated into a variable with an initial value of 0, called the weighted valid count. For example, if the first subsequent batch is valid, the weighted valid count is increased by 10 and divided by 55; if the second subsequent batch is also valid, it is increased by 9 and divided by 55, and so on.

[0041] The absolute number of valid batches is counted, i.e., the total number of batches recorded as valid, without distinguishing weights. For example, if 7 out of 10 subsequent batches have a probability migration distance of less than 0.15, then the absolute number of valid batches is 7.

[0042] Two thresholds are set: a cumulative threshold of 0.75 for the weighted effective count and a minimum threshold of 6 for the number of valid batches. When the weighted effective count exceeds 0.75 and the absolute number of valid batches reaches or exceeds 6, the cumulative number of valid batches is deemed to meet the upgrade criteria. Both conditions must be met simultaneously for the upgrade to pass. If only one condition is met while the other is not, for example, if the weighted effective count reaches 0.8 but the absolute number of valid batches is only 5, the upgrade criteria are not met.

[0043] When the upgrade conditions are met, the temporary mapping assumption established in step S4 is upgraded to a permanently effective mapping rule. The specific upgrade operation is as follows: the exchange operation type and interface identifier in the temporary mapping assumption are written into the permanent configuration table. Subsequent data streams from the same data source will automatically perform field exchange, eliminating the need for repeated checks. Simultaneously, the monitoring status corresponding to the temporary mapping assumption is cleared, and the upgrade timestamp is recorded.

[0044] When the upgrade conditions are not met—that is, the weighted effective count is less than or equal to 0.75, or the absolute number of effective batches is less than 6—a re-evaluation or transfer to manual verification is triggered. Specifically, there are two scenarios: If the absolute number of effective batches is 0, it means all subsequent batches fail, and the temporary mapping hypothesis is discarded, with subsequent batches restarting the complete detection process. If the absolute number of effective batches is between 1 and 5, or the weighted effective count is less than 0.75, the temporary mapping hypothesis, along with the probability migration distance calculation results for all subsequent batches, is packaged into a pending review record and pushed to the platform's manual verification interface. Operators then determine whether to ultimately adopt the exchange rule. After manual verification, the decision is made based on the verification results to either write the temporary mapping hypothesis to the permanent configuration table or completely delete it.

[0045] The working principle of this invention is as follows: Continuous data streams from multiple heterogeneous medical data sources are acquired in real time. Each data stream contains paired blood pressure value fields, and a high-quality reference data distribution provided by a verified interface is obtained. For each pair of blood pressure value fields in the current batch, the likelihood matching degree of the original combination and the exchanged combination under the reference distribution is calculated using discretized equal probability regions and hierarchical index ratios, constructing a likelihood ratio set. Based on the median deviation and directional consistency of this set, it is determined whether there are any field inversion anomalies. Simultaneously, probability migration distance is used to calculate and correct the differences and compatibility between the corrected data and the reference distribution. The system compares the tolerance upper limit; when the median deviates from the threshold and the probability migration distance is lower than the tolerance upper limit, it automatically performs field swapping to generate corrected data, and reorganizes the corrected data stream according to the implicit order identifier and prime number mapping multiplication and remainder method. At the same time, it establishes a temporary mapping hypothesis with the interface identifier and time window as the joint primary key; it continuously monitors multiple subsequent batches, assigns decreasing weights to each batch, and counts the weighted effective count and the number of effective batches. When both exceed the preset threshold, the temporary hypothesis is upgraded to a permanently effective mapping rule; otherwise, it is re-judged or transferred to manual verification, thereby achieving closed-loop collaborative optimization of data governance and intelligent analysis.

[0046] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A collaborative optimization method based on medical data resource governance and intelligent analysis, characterized in that, Includes the following steps: S1: Real-time acquisition of continuous data streams from multiple heterogeneous medical data sources, each of which includes a pair of blood pressure value fields, while also acquiring high-quality reference data distributions provided by a verified interface; S2: For each pair of blood pressure value fields in the current batch, calculate the likelihood matching degree of the original field combination and the exchanged field combination under the reference data distribution, and construct a set of likelihood ratios based on the matching degree difference of all samples. S3: Based on the median deviation of the likelihood ratio set and the consistency of the deviation direction of each sample, determine whether there is a field inversion anomaly in the current batch. At the same time, use probability migration distance to calculate the difference between the corrected data distribution and the reference data distribution, and determine whether the corresponding difference is lower than the tolerance limit. S4: When the median deviates beyond the threshold and the probability migration distance is below the tolerance limit, the field swap operation is automatically executed to generate corrected data, the corrected data is output to the subsequent analysis process, and a temporary mapping hypothesis record is established to record the corresponding swap relationship; S5: Continuously monitor the continued validity of the temporary mapping hypothesis and the probability migration distance of the corrected data in each batch in subsequent batches. When the cumulative number of valid batches exceeds the preset threshold and the migration distance is always lower than the tolerance limit, upgrade the temporary mapping hypothesis to a permanently effective mapping rule; otherwise, trigger a re-judgment or switch to the manual verification path.

2. The method for collaborative optimization based on medical data resource governance and intelligent analysis according to claim 1, characterized in that, S2 specifically includes: Discretize the continuous probability density of the reference data distribution into multiple equally probable regions; For each pair of blood pressure value fields in the current batch, locate the density level of the original field combination and the exchanged field combination falling into the equally probable region, and calculate the density ratio of the two combinations in the same level. After taking the natural logarithm of the density ratios of all samples in the corresponding batch, sort them by numerical value and select a preset number of logarithmic values ​​in the middle position as the likelihood ratio set.

3. The method for collaborative optimization based on medical data resource governance and intelligent analysis according to claim 2, characterized in that, The calculation of the density ratio of the two combinations at the same level specifically includes: The two-dimensional space of the reference data distribution is divided at equal intervals according to the cumulative probability of each dimension, forming multiple rectangular cells with equal probability areas, and each cell is pre-assigned a hierarchical number; Determine the rectangular cells where the original field combination and the swapped field combination fall respectively, and read the corresponding two hierarchical numbers; The ratio of the smaller value to the larger value of the two hierarchical indices is taken as the density ratio of the corresponding samples.

4. The method for collaborative optimization of medical data resource governance and intelligent analysis according to claim 1, characterized in that, S3 specifically includes: The relative deviation index is obtained by calculating the ratio of the median of the likelihood ratio set to the absolute median difference of the likelihood ratio set. The ratio of the difference between the number of samples with a value greater than 0 and the number of samples with a value less than 0 in the statistical likelihood ratio set is taken as the directional consistency coefficient. When the relative deviation index exceeds the preset threshold and the directional consistency coefficient is higher than the effective threshold, it is determined that there is a field inversion anomaly in the current batch.

5. The method for collaborative optimization based on medical data resource governance and intelligent analysis according to claim 1, characterized in that, S4 specifically includes: Read the original order of each pair of blood pressure value fields in the current batch, swap the positions of the first and second fields, and generate the swapped field pair; The swapped field pairs are reassembled into a corrected data stream in the original record order, and the corrected data stream is passed to the subsequent analysis stage; Using the interface identifier of the current data source and the current time window as a composite primary key, record a temporary mapping hypothesis that includes the exchange operation type and the effective monitoring period length.

6. The method for collaborative optimization of medical data resource governance and intelligent analysis according to claim 5, characterized in that, The process of reassembling the swapped field pairs into a corrected data stream according to the original record order specifically includes: Extract the implicit sequence identifier attached to each original record in the current batch. The implicit sequence identifier is the numerical value of the arrival order of the record in the data stream. The modified field obtained by exchanging fields in each record is bound to the implicit sequence identifier of the records to form an ordered entry; The data is rearranged in ascending order based on the numerical values ​​of the implicit sequence identifiers in all ordered entries. The rearranged correction field pairs are then concatenated sequentially to obtain the corrected data stream.

7. The method for collaborative optimization of medical data resource governance and intelligent analysis according to claim 6, characterized in that, The step of binding the modified field obtained by exchanging fields in each record with the implicit sequence identifier of the records to form an ordered entry specifically includes: A prime number mapping table is pre-defined, which corresponds one-to-one with the value space of the implicit sequence identifier. The corresponding prime number is found based on the implicit sequence identifier of the current record. The first value in the correction field pair is multiplied by the found prime number, and the last six digits of the product are taken as the first binding value. The second value in the correction field pair is multiplied by the same prime number, and the last six digits of the product are taken as the second binding value. The implicit sequence identifier, the first binding value, and the second binding value are concatenated into an ordered entry in chronological order.

8. The method for collaborative optimization of medical data resource governance and intelligent analysis according to claim 1, characterized in that, S5 specifically includes: Assign a weight value to each subsequent batch that decreases as the batch number increases, and normalize the sum of all weight values ​​to a unit of 1; Determine whether the probability migration distance of each subsequent batch is lower than the tolerance limit. Record the batches with the result lower as valid batches and add the weight value corresponding to the valid batches to the weighted valid count. When the weighted valid count exceeds the preset cumulative threshold, and the absolute number of valid batches also reaches the minimum number threshold, the cumulative number of valid batches is determined to meet the upgrade conditions.

Images