DRGs-based clinical medication data management system

Abstract

The invention relates to the technical field of medical information, and discloses a clinical medication data management system based on DRGs. The system comprises a data acquisition module, a mode construction module, a deviation identification module, a cause association module and a strategy generation module. According to the system, medicine use details and medicine use time track data under each DRG group are extracted and integrated, and a typical medicine use mode map is constructed to serve as a behavior reference. In real-time monitoring, the system performs multi-dimensional comparison on new case data and an atlas, and identifies drug use sequence, dosage and cost abnormity. The system performs intelligent attribution on the deviation based on a clinical medication knowledge graph, traverses drug interaction, treatment specifications and cost influence paths, forms a structured cause association network, and generates medication adjustment suggestions according to matching. According to the scheme, closed-loop management from clinical medication behavior modeling to deviation depth analysis and intervention strategy generation is realized, and medication monitoring fineness and decision support effectiveness are improved.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, specifically to a clinical medication data management system based on DRGs. Background Technology

[0002] Under the DRG-based medical management model, refined control over clinical medication use has become a core requirement for medical institutions. Existing solutions primarily rely on statistical analysis and threshold monitoring of historical cost data, focusing on calculating macro-level indicators such as total drug costs, average costs, or drug ratios for each DRG group. While these methods can achieve traceability of financial results and early warning of exceeding limits, their management logic remains at the level of total costs, failing to delve into the internal processes of clinical medication use.

[0003] The limitation of existing technologies lies in their lack of quantitative definitions and descriptions of standardized and rational medication use patterns under specific DRG groupings. Because they cannot establish clinical behavioral benchmarks that integrate drug selection, dosage, and timing, the system struggles to distinguish between reasonable cost fluctuations due to the complexity of the condition and resource consumption caused by irregularities such as disordered medication sequence or inappropriate dosage. Consequently, management practices are lagging and lack specificity.

[0004] When the system detects abnormal costs or non-compliant drug use, existing solutions typically trigger alerts based on pre-defined, isolated single rules. The limitation of this approach is that it cannot systematically trace and present the multiple intertwined potential causes behind a single medication deviation. Decision support thus remains at the level of simply informing about surface phenomena, lacking in-depth analysis and explanation of the correlation between the deviation's root causes. Summary of the Invention

[0005] The purpose of this invention is to provide a clinical medication data management system based on DRGs to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a clinical medication data management system based on DRGs, the system comprising:

[0007] The data acquisition module is used to extract a set of structured medication records corresponding to each group from the electronic medical record database of the target medical institution according to the diagnosis-related group identifier. The set of structured medication records includes detailed drug usage data, medication time trajectory data, and associated medical expense item data.

[0008] The pattern construction module is used to jointly analyze the drug usage details data and medication time trajectory data to construct a typical medication pattern map for each case group;

[0009] The deviation identification module is used to compare the real-time collected new case medication data with the typical medication pattern map in multiple dimensions, identify medication order deviation, dosage abnormal deviation and cost increase abnormal deviation, and generate a medication deviation report.

[0010] The cause association module is used to generate a deviation cause association network by traversing the drug interaction paths, treatment standard paths, and cost influencing factor paths associated with the deviation items in a preset clinical drug knowledge graph based on the medication deviation report.

[0011] The strategy generation module is used to retrieve a preset medication optimization rule base based on the deviation cause association network, match and obtain a set of medication adjustment suggestions for specific deviation types, and send the set of medication adjustment suggestions to the clinical decision support terminal.

[0012] Preferably, the pattern construction module performs joint analysis on the drug usage details data and medication time trajectory data to construct a typical medication pattern map, including:

[0013] The typical medication pattern diagram includes the order of drug use, dosage adjustment points, and cost accumulation trajectory.

[0014] The drug usage details data are processed by drug classification code mapping, which maps the names of various drugs to standard drug classification codes, forming a coded drug usage sequence.

[0015] The medication time trajectory data is divided into time windows, and the entire medication time trajectory is divided into multiple consecutive treatment stage time windows according to the treatment stage.

[0016] Within each treatment phase time window, the encoded drug use sequence is subjected to frequent pattern mining to identify drug combination patterns that frequently co-occur within the treatment phase time window, which are then used as typical drug combinations for the treatment phase.

[0017] Based on the chronological order of the time windows, the typical drug combinations identified in each treatment stage are connected according to the time flow, and the addition, deletion and changes in drug use between stages are marked on the connection edges to form a preliminary directed graph of medication patterns.

[0018] Extract drug cost data for each treatment stage time window from the medical cost item data, calculate the cost contribution of the typical drug combination, and label the cost contribution as a weight attribute on the corresponding node in the directed graph of the medication pattern to complete the construction of the typical medication pattern graph.

[0019] Preferably, the deviation identification module performs multi-dimensional comparison between the real-time collected new case medication data and the typical medication pattern atlas to identify deviations, including:

[0020] By analyzing the medication data of the new cases, the drug use sequence and medication time trajectory of the new cases were obtained;

[0021] Align and match the medication time trajectory of the new case with the treatment stage time window in the typical medication pattern atlas to determine the current treatment stage of the new case;

[0022] Obtain the typical drug combination corresponding to the current treatment stage in the typical drug use pattern map, compare the drug use sequence of the new case with the typical drug combination, and if the drug use sequence of the new case contains drugs from the atypical drug combination, it is recorded as a drug use order deviation.

[0023] Calculate the actual dosage of each drug used in the current treatment phase of the new case and compare it with the preset standard dosage range for the treatment phase. If the actual dosage exceeds the standard dosage range, it is recorded as an abnormal dosage deviation.

[0024] The total cost of medication for the new case from the start of treatment to the current stage is accumulated and compared with the historical average cost at the same treatment stage node in the typical medication pattern graph. If the current total cost exceeds the preset percentage threshold of the historical average cost, it is recorded as an abnormal deviation in cost growth.

[0025] By integrating the medication sequence deviation, dosage abnormality deviation, and cost increase abnormality deviation, a structured medication deviation report is generated.

[0026] Preferably, the causal association module generates a causal association network for deviations by traversing a preset clinical medication knowledge graph based on the medication deviation report, including:

[0027] Extract the atypical drugs corresponding to the medication order deviation, the specific drugs and excess values ​​corresponding to the dosage abnormality deviation, and the cost overrun drug categories corresponding to the cost growth abnormality deviation from the medication deviation report.

[0028] Using the aforementioned atypical drugs, specific drugs, and drug categories with excessive costs as query entities, entity retrieval is performed in the clinical drug knowledge graph to locate the node position of the query entity in the knowledge graph.

[0029] Read all preset types of relationship paths starting from nodes from the clinical drug knowledge graph. The preset types of relationship paths include drug substitution relationship paths, drug incompatibility relationship paths, treatment guideline recommendation relationship paths, and drug cost-effectiveness relationship paths.

[0030] Perform a multi-hop traversal along each read relationship path, collect all associated intermediate entity nodes and relationship edges on the path, and form multiple subgraphs with the query entity as the root node.

[0031] All collected subgraphs are merged, duplicate nodes and edges are removed, and a network is constructed that connects the various deviations in the medication deviation report with potential clinical and cost factors to establish the causes of the deviations.

[0032] Preferably, the strategy generation module retrieves a set of medication adjustment suggestions from a preset medication optimization rule base based on the deviation cause association network, including:

[0033] The deviation cause correlation network is structurally decomposed into several independent deviation factor graphs. Each deviation factor graph describes a major deviation type and its associated causal path.

[0034] For each deviation factor graph, extract its core deviation type and key causal entity, and combine the deviation type and key causal entity into a query statement.

[0035] The query statement is used to perform pattern matching in the medication optimization rule base, which stores rules in the format of "deviation condition-cause condition-adjustment suggestion";

[0036] When the deviation type and key causal entity in the query statement match the deviation condition and causal condition in a rule, the rule is triggered and the corresponding adjustment suggestion is obtained.

[0037] The adjustment suggestions obtained from all deviation factor plots will be summarized and deduplicated to form a set of medication adjustment suggestions for the new case.

[0038] Preferably, the strategy generation module sends the set of medication adjustment suggestions to the clinical decision support terminal, including:

[0039] Each medication adjustment suggestion in the set of suggestions is prioritized, and the prioritization is calculated based on the severity of the impact of the deviation associated with the adjustment suggestion on treatment safety and cost.

[0040] For each adjustment suggestion, a corresponding execution context description is generated, which includes the applicable treatment stage, the drugs involved, a summary of the expected adjustment effect, and reference basis;

[0041] The sorted medication adjustment recommendations and their execution context descriptions are encapsulated into a decision support message in a specific format;

[0042] The decision support message is pushed to the clinical decision support terminal of the target clinician through the medical data exchange interface and highlighted in a designated area of ​​the terminal interface.

[0043] Preferably, within each treatment phase time window, the coded drug use sequence undergoes frequent pattern mining processing to identify drug combination patterns that frequently co-occur within the treatment phase time window, which are then used as typical drug combinations for the treatment phase. This includes:

[0044] Obtain coded drug use sequences for all cases within the current treatment phase time window, with each sequence containing multiple standard drug classification codes sorted by drug use time;

[0045] Set a minimum support threshold, which represents the minimum frequency at which the drug combination appears in all case sequences;

[0046] A frequent pattern growth algorithm is used to traverse all drug usage sequences, count the frequency of each drug code, and generate a header table by sorting them in descending order of frequency.

[0047] Based on the head table, the drug use sequence of each case is projected to construct a conditional pattern base;

[0048] Recursively mine frequent itemsets on the conditional pattern base. When the support of an itemset reaches the minimum support threshold, the drug combination is marked as a high-frequency combination.

[0049] All identified high-frequency drug combinations are sorted in descending order of support, and the drug combinations with the highest support are selected as typical drug combinations for the current treatment phase time window.

[0050] Typical drug combinations are bound and stored with corresponding treatment phase time windows for subsequent construction of a directed graph of medication patterns.

[0051] Preferably, aligning and matching the medication time trajectory of the new case with the treatment stage time window in the typical medication pattern atlas to determine the current treatment stage of the new case includes:

[0052] Analyze the medication timeline of new cases, extracting the start time of medication, the time of each drug's use, and the current time.

[0053] Predefined treatment phase time window division rules are obtained from typical medication pattern diagrams, the rules including the duration range and sequential relationship of each treatment phase;

[0054] The starting time of medication for new cases is taken as the zero point of the time axis, and the offset of each medication usage time point relative to the zero point is calculated.

[0055] Based on the duration range of the treatment phase time window, multiple consecutive treatment phase time windows are divided on the time axis.

[0056] The drug use time sequence of new cases was matched with the time windows of the defined treatment stages, and the number of drug use events occurring in each time window was counted.

[0057] The treatment phase with the highest number of drug use events was selected as the most likely current treatment phase for new cases.

[0058] When there are multiple time windows with a similar number of drug use events, the time window with the highest time density is selected as the current treatment stage, taking into account the time density distribution characteristics of drug use.

[0059] The final identified treatment stage identifier is output for subsequent drug use sequence alignment analysis.

[0060] Preferably, the system further includes a medication pattern update module, used to dynamically update the typical medication pattern map based on feedback data from new cases, including:

[0061] Receive feedback data from the clinical decision support terminal, the feedback data recording the medication regimens actually adopted by doctors for new cases and the implementation status of the set of medication adjustment suggestions;

[0062] Extract proven effective medication pattern fragments from adopted medication regimens;

[0063] The similarity between the medication pattern fragment and the corresponding treatment stage node in the typical medication pattern atlas is calculated. If the similarity is higher than a preset threshold, the weight of the node in the atlas is strengthened. If the similarity is lower than the preset threshold but the medication result is well evaluated, the medication pattern fragment is added to the atlas as a new candidate pattern branch.

[0064] Based on feedback data from all new cases within a time period, the weights and branch structures of each node in the typical medication pattern atlas are recalculated and optimized to generate an updated version of the typical medication pattern atlas.

[0065] Preferably, the medication pattern update module adds medication pattern fragments as new candidate pattern branches to the atlas, including:

[0066] Analyze the differences between the medication pattern segment and existing atlas nodes in terms of drug composition, medication sequence and dosage, and locate the target node in the typical medication pattern atlas that is the same as the treatment stage of the medication pattern segment;

[0067] Using the target node as the parent node, create new child nodes to represent the medication mode segment;

[0068] Establish a directed edge between the parent node and the new child node, and label the specific differences between the parent node and the new child node on the directed edge;

[0069] Initialize weights for new child nodes, which are calculated based on the number of cases that generated this medication pattern segment and the medication outcome assessment score.

[0070] Compared with the prior art, the beneficial effects of the present invention are:

[0071] The system integrates and models discrete static information such as drug names, specifications, and dosages with dynamic sequence information such as dosing start time, intervals, and duration. Through data mining methods, the system can characterize the general and representative medication behavior structure under specific disease groups. This structured behavioral data model provides an objective and detailed benchmark for subsequent real-time monitoring, shifting the management focus from financial outcome accounting to the standardized assessment of clinical behavior processes.

[0072] Based on a pre-defined clinical medication knowledge graph, the system traverses multiple logical paths associated with identified deviations, generating a structured network of deviation causal relationships. After identifying a specific medication deviation, the system, according to pre-defined entity relationships and rules in the knowledge graph, simultaneously activates interaction paths between drugs, normative paths in clinical treatment guidelines, and paths related to factors affecting total treatment costs. Through path traversal and association, the system automatically constructs a network analysis structure centered on the current deviation and connecting multiple potential impact dimensions. This process achieves a three-dimensional attribution of a single anomaly.

[0073] The establishment of typical medication use pattern maps enables the system to perform sequence matching and compliance calculations on real-time medication behaviors, thereby identifying abnormal medication sequences and dosages that deviate from the established clinical pathways and providing early warnings of potential cost increases. This changes the traditional management model that can only compare total costs after the fact. Through causal association networks, the system outputs not isolated alarm messages, but a set of potential causes with inherent logical connections. This network reveals the possible root causes of deviations in different dimensions such as pharmacological safety, treatment standardization, and economic rationality, allowing subsequent interventions to be based on more comprehensive causal inferences rather than single-dimensional judgments. Attached Figure Description

[0074] Figure 1 This is a timing diagram of the DRG-based clinical medication data management system described in this invention.

[0075] Figure 2 A flowchart for constructing a typical drug use pattern map;

[0076] Figure 3 A flowchart generated from the bias cause association network;

[0077] Figure 4 A biaxial analysis graph showing the frequency and impact scores of medication deviation types;

[0078] Figure 5 A bar chart comparing the number of drugs in different DRG groups. Detailed Implementation

[0079] 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.

[0080] Please see Figure 1 This invention provides a clinical medication data management system based on DRGs. The system includes: a data acquisition module that extracts a set of structured medication records corresponding to each group from the electronic medical record database of the target medical institution, based on the diagnosis-related group identifier. This set includes detailed drug usage data, medication time trajectory data, and associated medical expense item data. A pattern construction module performs joint analysis on the collected detailed drug usage data and medication time trajectory data to construct a typical medication pattern map for each case group. A deviation identification module performs multi-dimensional comparison between the real-time collected new case medication data and the constructed typical medication pattern map to identify various deviations, including medication sequence deviations, dosage abnormalities, and cost increase abnormalities, and generates a structured medication deviation report. A causal association module, based on the generated medication deviation report, traverses a preset clinical medication knowledge graph to find drug interaction paths, treatment standard paths, and cost influencing factor paths associated with the deviation items in the report, thereby generating a deviation causal association network connecting deviations and potential causes. The strategy generation module retrieves a preset medication optimization rule base based on the deviation cause association network, obtains a set of medication adjustment suggestions for specific deviation types through matching, and sends the set to the clinical decision support terminal to provide reference for clinical decision-making.

[0081] In one embodiment of the present invention, see [reference] Figure 2The pattern construction module performs the process of constructing a typical medication use pattern atlas. The typical medication use pattern atlas includes the order of drug use, dosage adjustment nodes, and cost accumulation trajectory. The pattern construction module performs drug classification code mapping processing on the drug use details data, mapping the various drug brand names or generic names recorded in the electronic medical record to a unified standard drug classification code. For example, "Ceftriaxone Sodium for Injection" and "Cefazolin Sodium" are mapped to the "J01D-Third Generation Cephalosporin" classification code. This processing forms a coded drug use sequence, and each sequence consists of a series of standard drug classification codes ordered by time. The pattern construction module performs time window division processing on the medication time trajectory data. Based on the clinical treatment path of the target diagnosis related group, the complete inpatient medication time trajectory is divided into continuous treatment stage time windows with clear clinical significance, such as the preoperative preparation stage time window, the surgical stage time window, and the postoperative recovery stage time window.

[0082] Within each treatment phase time window, the pattern construction module performs frequent pattern mining on the coded drug use sequences to identify high-frequency co-occurring drug combination patterns as typical drug combinations for that treatment phase. Frequent pattern mining obtains a set of coded drug use sequences for all historical cases within the current treatment phase time window. A minimum support threshold is set to define the standard for "high frequency." A frequent pattern growth algorithm iterates through the sequence set, counting the frequency of each drug classification code and generating a head table sorted in descending order of frequency. Based on the head table, the drug use sequence for each case is projected to construct a conditional pattern base. Frequent itemsets satisfying the minimum support threshold are recursively mined from the conditional pattern base. For a drug combination pattern, its support within the current treatment phase time window is calculated using the following formula:

[0083]

[0084] in: Indicates support level, This indicates the number of cases involving this drug combination. This indicates the total number of cases within the current treatment phase time window, when the support of the drug combination... When a drug combination reaches or exceeds a preset minimum support threshold, it is marked as a high-frequency drug combination. All identified high-frequency drug combinations are then ranked according to their support. Sort the drugs in descending order and select the top K drug combinations with the highest support as the typical drug combinations for the current treatment phase time window. Then, bind and store the typical drug combinations with their corresponding treatment phase time window identifiers.

[0085] Based on the chronological order of the treatment phase time windows, the pattern construction module connects the typical drug combinations identified in each treatment phase according to the time flow. This connection operation constructs a directed graph structure, where nodes represent typical drug combinations within a specific treatment phase time window. Changes in drug usage between phases are marked on the connection edges; for example, from the pre-operative phase to the surgical phase, newly added drug classification codes and discontinued drug classification codes are marked. This marking forms a preliminary directed graph of medication patterns. Drug cost data related to typical drug combinations within each treatment phase time window is extracted from the medical cost item data. The cost contribution of each typical drug combination is calculated, defined as the proportion of the total cost of drugs included in that typical drug combination to the total cost of all drugs within the current treatment phase time window. The pattern construction module uses the calculated cost contribution as a weight attribute and marks it on the node representing that typical drug combination in the directed graph of medication patterns, thus completing the construction of the typical medication pattern atlas. In some embodiments, the cost contribution weight is visualized, for example, by the size or color intensity of the node.

[0086] It is understandable that the treatment stage definition used for time window segmentation is derived from clinical pathways or expert consensus. Optionally, the minimum support threshold and the number K of selected typical drug combinations can be configured differently based on the characteristics of different diagnosis-related groupings. In some embodiments, the standard drug classification code adopts the National Reimbursement Drug List classification code or the ATC classification system code. It is understandable that the coded drug use sequence eliminates the impact of differences in drug name descriptions on the analysis. Optionally, the directed graph of medication patterns allows for the existence of multiple typical drug combination nodes corresponding to a treatment stage time window, to represent the different medication strategy branches existing in that stage.

[0087] In one embodiment of the present invention, see [reference] Figure 3The deviation identification module performs multi-dimensional comparisons between the real-time collected medication data of new cases and the constructed typical medication pattern atlas to identify deviations. The module analyzes the medication data of new cases to obtain the drug usage sequence and the medication time trajectory of each case. The drug usage sequence includes a mapped list of standard drug classification codes, and the medication time trajectory includes the precise timestamp of each drug's use. The deviation identification module aligns and matches the medication time trajectory of new cases with the treatment stage time windows in the typical medication pattern atlas. This alignment and matching process calculates based on the preset start time offset and duration of each treatment stage time window. By comparing which treatment stage time window the offset of the new case's current time point relative to its medication start point falls within, the module determines the current treatment stage identifier of the new case. Obtain the typical drug combinations corresponding to the current treatment stage from the typical drug use pattern map. A typical drug combination is a set of several frequently occurring standard drug classification codes. Compare the drug use sequence of the new case with the typical drug combinations. The comparison operation checks whether each code in the drug use sequence of the new case exists in the typical drug combination code set. If the drug use sequence of the new case contains drugs from atypical drug combinations, that is, at least one code is not in the typical drug combination set, it is recorded as a drug use order deviation.

[0088] The deviation identification module calculates the actual dosage of each drug used in the current treatment phase of a new case. The actual dosage is derived from the prescription details in the new case's medication data and compared with the preset standard dosage range for the treatment phase. The standard dosage range for the treatment phase is derived from drug instructions, clinical guidelines, or historical data statistics. For each drug, if the actual dosage is less than the lower limit or greater than the upper limit of the standard dosage range, it is recorded as an abnormal dosage deviation for that drug. The deviation identification module accumulates the total drug cost for the new case from the start of treatment to the current phase. The accumulation process sums all incurred drug costs in the new case's medical expense data and compares it with the historical average cost at the same treatment phase node in the typical medication pattern graph. The historical average cost is based on the statistical average of drug costs incurred by a large number of past cases at the same treatment phase. A preset percentage threshold is used as the judgment standard during the comparison. If the current total cost exceeds the historical average cost by a degree greater than or equal to the preset percentage threshold, it is recorded as an abnormal cost growth deviation. The deviation identification module integrates the identified deviations in medication order, dosage, and cost increase. Each deviation record includes the deviation type, the drug identifier involved, and the specific value or description of the deviation, generating and outputting a structured medication deviation report.

[0089] In practice, medication deviation reports are input into the causal association module, which generates a deviation causal association network by traversing a pre-defined clinical medication knowledge graph based on the reports. The module extracts keyword information from the reports, including atypical drug names or codes corresponding to medication order deviations, specific drug names or codes and excess dosages corresponding to dosage abnormalities, and cost-overrun drug categories corresponding to cost-overrun deviations. Using the extracted atypical drugs, specific drugs, and cost-overrun drug categories as query entities, entity retrieval is performed in the clinical medication knowledge graph. Entity retrieval is accomplished by matching the attribute values ​​of nodes in the knowledge graph to locate the query entity's node position. All pre-defined types of relationship paths originating from these located nodes are read from the clinical medication knowledge graph. These pre-defined relationship paths are defined during system initialization and include drug substitution relationship paths, drug incompatibility relationship paths, treatment guideline recommendation relationship paths, and drug cost-effectiveness relationship paths. A multi-hop traversal is performed along each read relational path. Following the path's direction, the traversal sequentially visits other entity nodes directly or indirectly connected to the starting node, collecting all associated intermediate entity nodes and relational edges along the path, forming multiple subgraph structures with the query entity as the root node. All collected subgraphs are merged, and the merging operation identifies and removes duplicate nodes and edges, constructing a deviation causal association network connecting various deviations in the medication deviation report with potential clinical and cost factors. In some embodiments, to quantify the association strength, the number of hops in the relational path traversal is recorded, and a decay factor is defined. The formula for calculating the association strength weight from a deviation node to a causal node in the deviation causal association network is:

[0090]

[0091] in: The weight representing the association strength between the deviation node and the causal node. This represents the set of all paths connecting the deviation node and the cause node. Representing a path The number of jumps, It is a decay factor and Understandably, this weight is used to assess the importance of different causes during subsequent strategy generation. The cause association module outputs the completed deviation cause association network, which is stored in graph data format. Optionally, drug incompatibility relationship paths can be linked to entity nodes describing adverse interaction mechanisms, and drug cost-benefit relationship paths can be linked to entity nodes describing average daily treatment costs or pharmacoeconomic evaluation results.

[0092] In one embodiment of the present invention, the strategy generation module retrieves a set of medication adjustment suggestions from a preset medication optimization rule base based on the deviation cause association network. The strategy generation module performs structural decomposition processing on the deviation cause association network, splitting it into several independent deviation factor graphs. Each deviation factor graph describes a major deviation type and its associated causal path. For each deviation factor graph, the strategy generation module extracts its core deviation type and key causal entity, combines the deviation type and key causal entity into a query statement, and uses the query statement to perform pattern matching in the medication optimization rule base. The medication optimization rule base stores rules in the format of "deviation condition-causal condition-adjustment suggestion". When the deviation type and key causal entity in the query statement match the deviation condition and causal condition in a rule, the rule is triggered, and the corresponding adjustment suggestion is obtained. The strategy generation module summarizes and deduplicates the adjustment suggestions obtained from all deviation factor graphs to form a set of medication adjustment suggestions for new cases. In some embodiments, deviation types include medication sequence deviation, dosage abnormal deviation, or cost increase abnormal deviation. Optionally, key causal entities include specific drug nodes in a drug interaction pathway or guideline entry nodes in a treatment protocol pathway.

[0093] The strategy generation module sends a set of medication adjustment suggestions to the clinical decision support terminal. The module then prioritizes each suggestion in the set, calculating the priority based on the severity of the impact of the associated deviation on treatment safety and cost. The priority score is calculated using the following formula:

[0094]

[0095] in: This represents the priority score of the i-th adjustment suggestion. This indicates the level of impact of the deviation associated with the recommendation on treatment safety. This indicates the level of impact of the deviation associated with the recommendation on costs. and These are preset weighting coefficients. The strategy generation module generates a corresponding execution context description for each adjustment suggestion. The execution context description includes the applicable treatment stage, the drugs involved, a summary of the expected adjustment effect, and reference basis. The strategy generation module encapsulates the sorted medication adjustment suggestions and their execution context descriptions into a decision support message in a specific format. This message is pushed to the target clinician's clinical decision support terminal via a medical data exchange interface and highlighted in a designated area of ​​the terminal's interface. In some embodiments, the decision support message is encapsulated using the HL7FHIR standard format. Optionally, highlighting is achieved by changing the text color or adding visual identifiers. It is understood that the priority sorting process ensures that the most critical adjustment suggestions are presented to clinicians first. It is understood that the execution context description provides the clinical background and basis for the adjustment suggestions.

[0096] In one embodiment of the invention, frequent pattern mining is performed on the coded drug use sequences to identify typical drug combinations. This process obtains the coded drug use sequences of all historical cases within the current treatment phase time window. Each sequence contains multiple standard drug classification codes sorted by medication time. A minimum support threshold is set to define the high-frequency standards, and the minimum support threshold is a value between 0 and 100%. A frequent pattern growth algorithm is used to traverse all drug use sequences. The algorithm first counts the frequency of each drug classification code in the entire sequence set and generates a head table by sorting the sequences from high to low frequency. Then, a projection operation is performed on the drug use sequence of each case based on the head table to construct a corresponding conditional pattern base for each drug classification code. The conditional pattern base is a set that records all sequences containing that code and their subsequent items. Frequent itemsets are recursively mined on the conditional pattern base. For the currently processed drug classification code, the frequency of other codes is counted on its conditional pattern base to generate a new conditional pattern base, until no new frequent itemsets can be generated. When the support of a discovered drug combination itemset reaches or exceeds a preset minimum support threshold, the drug combination is marked as a high-frequency drug combination. The specific formula for calculating support is as follows:

[0097]

[0098] in Indicates the support of a drug combination itemset. This indicates the number of cases that include the drug combination item set. This represents the total number of cases within the current treatment phase time window. All identified high-frequency drug combinations are ranked according to their support. Sort in descending order and select support. The top few drug combinations are designated as typical drug combinations for the current treatment phase time window. These typical drug combinations are then linked and stored with their corresponding treatment phase time window identifiers. See Table 1 for statistics on the number of drug usage events within a single treatment phase time window.

[0099] Table 1: Statistical Table of Time Window Matching in Treatment Phases

[0100] Treatment phase time window marking Time window start offset (days) Time window end offset (days) Number of drug use events Preoperative preparation stage 0 2 15 Surgical period 2 4 8 Postoperative recovery stage 4 10 22

[0101] In practice, the medication timeline of new cases is aligned and matched with the treatment phase time windows in a typical medication pattern diagram. This alignment and matching process analyzes the medication timeline of new cases, extracting the medication start time, the time points of each drug use, and the current time point. Predefined treatment phase time window division rules are obtained from the typical medication pattern diagram. These rules include the duration range and sequential relationship of each treatment phase. The medication start time of the new case is used as the zero point of the time axis, and the offset of each drug use time point relative to the zero point is calculated, typically in hours or days. Based on the duration range of the treatment phase time windows, multiple consecutive treatment phase time windows are divided on the time axis. The sequence of drug use time points of new cases is time-series matched with the divided treatment phase time windows, and the number of drug use events occurring within each time window is counted. The treatment phase time window with the most drug use events is selected as the most likely treatment phase the new case is currently in. When multiple time windows have a similar number of drug use events, the time density distribution characteristics of drug use are considered for judgment. The calculation method is the ratio of the number of drug use events within a specific time window to the duration of that time window:

[0102]

[0103] in: Indicates time density, This indicates the number of drug use events within that time window. This indicates the duration of the time window. Select the time density. The highest time window is designated as the current treatment stage. The final determined treatment stage identifier is output. In some embodiments, the time window offset is calculated to the minute level to handle short-term intensive medication scenarios. Optionally, when medication use events occur near the time window boundaries, overlapping time windows or weighted counting can be used for matching to improve accuracy. It is understood that time density calculation helps distinguish stages with the same number of events but different medication rhythms. In some embodiments, the treatment stage time window division rules can be dynamically configured based on different diagnosis-related groupings. It is understood that the accuracy of alignment matching directly affects the effectiveness of subsequent deviation identification. Optionally, time windows where no medication use events have occurred are given lower priority during the matching stage.

[0104] See Figure 4 This is a biaxial chart analyzing the frequency and impact scores of different medication deviation types, illustrating the correlation between the frequency and severity of various deviations. Dosage abnormalities, possessing both "high frequency" and "high impact score," represent the highest-risk deviation type. While medication order deviations also have a high frequency, their impact is relatively low. The frequency of deviations and impact scores are not entirely positively correlated, indicating the need to monitor both "high-frequency, low-impact" and "low-frequency, high-impact" deviations. Dosage abnormalities are clearly identified as the primary intervention target (high-risk), requiring the prioritization of stricter medication monitoring rules; deviations involving abnormal cost increases require supplementary cost control strategies. More clinical review resources should be allocated to high-risk deviations (dosage abnormalities), while a lightweight management approach with automated early warning systems should be adopted for low-risk deviations (medication order).

[0105] In one embodiment of the present invention, the medication pattern update module dynamically updates the typical medication pattern atlas based on feedback data from new cases. The module receives feedback data from a clinical decision support terminal, which records the medication regimens actually adopted by doctors for new cases and the execution status of the medication adjustment suggestion set. It extracts validated medication pattern fragments from the actually adopted regimens; these fragments are drug usage sequences corresponding to specific treatment stages. The module calculates the similarity between the extracted medication pattern fragments and the nodes corresponding to the treatment stages in the typical medication pattern atlas. The similarity calculation is based on the consistency of the typical drug combinations represented by the medication pattern fragments and the atlas nodes in terms of drug composition and administration order. If the calculated similarity is higher than a preset similarity threshold, the weight of that node in the atlas is increased, typically by increasing the node's weight value. If the similarity is below a preset similarity threshold but the medication outcome is evaluated as good (meaning the case corresponding to the medication pattern segment meets preset positive evaluation criteria in terms of treatment effect, safety, and cost control), then the medication pattern segment is added as a new candidate pattern branch to the typical medication pattern atlas. Based on feedback data from all new cases within a time period, the medication pattern update module recalculates and optimizes the weights and branch structures of each node in the typical medication pattern atlas. The recalculation of weights comprehensively considers the node's historical weights, the number of times it has been recently strengthened, and the overall evaluation results of associated cases, generating an updated version of the typical medication pattern atlas.

[0106] The medication pattern update module adds medication pattern fragments as new candidate pattern branches to the atlas. The addition process analyzes the differences between the medication pattern fragment and existing atlas nodes in terms of drug composition, medication sequence, and dosage. It locates a target node in the typical medication pattern atlas that corresponds to the treatment stage of the medication pattern fragment. Using this target node as the parent node, new child nodes are created to represent the medication pattern fragment. A directed edge is established between the parent node and the new child node, and the specific difference attributes between the parent node and the new child node are labeled on the directed edge. These difference attributes can include newly added drug codes, deleted drug codes, or percentage dosage adjustments. Weights are initialized for the new child nodes. The initial weights are calculated based on the number of cases that generated this medication pattern fragment and the medication outcome evaluation score. The formula for calculating the initial weights is:

[0107]

[0108] in: This indicates the initial weight of the new child node. This indicates the number of new cases that generated this medication pattern fragment during the observation period. This represents a preset baseline value for the number of cases. This represents the average of the medication outcome assessment scores for these cases. This is a weighting coefficient used to balance the number of cases with the quality of outcomes. In some embodiments, the medication outcome assessment score is calculated comprehensively from multiple dimensions such as clinical outcome indicators, adverse drug reaction rate, and treatment cost. It is understood that the addition of new candidate pattern branches allows the typical medication pattern atlas to accommodate and characterize effective medication strategies that have been validated in practice but deviate from historically high-frequency patterns. In some embodiments, newly added sub-nodes, if their weight values ​​continue to increase and exceed a certain threshold in subsequent case comparisons, may be promoted to the main typical node of their respective treatment stage in future iterations. Optionally, the difference attributes labeled on the connection edges provide clinicians with intuitive information to quickly understand the differences between new branches and traditional patterns. It is understood that the dynamic update mechanism ensures that the typical medication pattern atlas can continuously evolve with the development of clinical practice.

[0109] See Figure 5 This is a bar chart comparing the number of drugs in different DRG groups, used to show the differences in drug usage scale among various diagnosis-related groups. The number of drugs in the surgical DRG group is generally higher than that in internal medicine and pediatrics, reflecting the higher complexity of drug use in surgical treatment. The number of drugs in internal medicine DRG2 (12) is significantly higher than that in internal medicine DRG1 (8), indicating that there are significant differences in drug needs among different diagnostic groups in the same department. This provides basic data for the "typical drug use pattern map" of different DRG groups, improving the accuracy of pattern matching. More pharmaceutical review resources can be matched for surgical DRGs (especially DRG2), and a simpler drug management process can be adopted for pediatric DRGs to improve resource utilization efficiency. DRG groups with a large number of drugs (such as surgical DRG2) are the key targets for cost control, and the drug combination of this group can be optimized in a targeted manner to reduce costs.

[0110] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0111] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A clinical medication data management system based on DRGs, characterized in that, The system includes: The data acquisition module is used to extract a set of structured medication records corresponding to each group from the electronic medical record database of the target medical institution according to the diagnosis-related group identifier. The set of structured medication records includes detailed drug usage data, medication time trajectory data, and associated medical expense item data. The pattern construction module is used to jointly analyze the drug usage details data and medication time trajectory data to construct a typical medication pattern map for each case group; The deviation identification module is used to compare the real-time collected new case medication data with the typical medication pattern map in multiple dimensions, identify medication order deviation, dosage abnormal deviation and cost increase abnormal deviation, and generate a medication deviation report. The cause association module is used to generate a deviation cause association network by traversing the drug interaction paths, treatment standard paths, and cost influencing factor paths associated with the deviation items in a preset clinical drug knowledge graph based on the medication deviation report. The strategy generation module is used to retrieve a preset medication optimization rule base based on the deviation cause association network, match and obtain a set of medication adjustment suggestions for specific deviation types, and send the set of medication adjustment suggestions to the clinical decision support terminal.

2. The clinical medication data management system based on DRGs according to claim 1, characterized in that, The pattern construction module performs joint analysis on the drug usage details data and medication time trajectory data to construct a typical medication pattern map, including: The typical medication pattern diagram includes the order of drug use, dosage adjustment points, and cost accumulation trajectory. The drug usage details data are processed by drug classification code mapping, which maps the names of various drugs to standard drug classification codes, forming a coded drug usage sequence. The medication time trajectory data is divided into time windows, and the entire medication time trajectory is divided into multiple consecutive treatment stage time windows according to the treatment stage. Within each treatment phase time window, the encoded drug use sequence is subjected to frequent pattern mining to identify drug combination patterns that frequently co-occur within the treatment phase time window, which are then used as typical drug combinations for the treatment phase. Based on the chronological order of the time windows, the typical drug combinations identified in each treatment stage are connected according to the time flow, and the addition, deletion and changes in drug use between stages are marked on the connection edges to form a preliminary directed graph of drug use patterns. Extract drug cost data for each treatment stage time window from the medical cost item data, calculate the cost contribution of the typical drug combination, and label the cost contribution as a weight attribute on the corresponding node in the directed graph of the medication pattern to complete the construction of the typical medication pattern graph.

3. The clinical medication data management system based on DRGs according to claim 1, characterized in that, The deviation identification module compares the real-time collected medication data of new cases with the typical medication pattern atlas in multiple dimensions to identify deviations, including: By analyzing the medication data of the new cases, the drug use sequence and medication time trajectory of the new cases were obtained; Align and match the medication time trajectory of the new case with the treatment stage time window in the typical medication pattern atlas to determine the current treatment stage of the new case; Obtain the typical drug combination corresponding to the current treatment stage in the typical drug use pattern map, compare the drug use sequence of the new case with the typical drug combination, and if the drug use sequence of the new case contains drugs from the atypical drug combination, it is recorded as a drug use order deviation. Calculate the actual dosage of each drug used in the current treatment phase of the new case and compare it with the preset standard dosage range for the treatment phase. If the actual dosage exceeds the standard dosage range, it is recorded as an abnormal dosage deviation. The total cost of medication for the new case from the start of treatment to the current stage is accumulated and compared with the historical average cost at the same treatment stage node in the typical medication pattern graph. If the current total cost exceeds the preset percentage threshold of the historical average cost, it is recorded as an abnormal deviation in cost growth. By integrating the medication sequence deviation, dosage abnormality deviation, and cost increase abnormality deviation, a structured medication deviation report is generated.

4. The DRG-based clinical medication data management system according to claim 3, characterized in that, The causal association module generates a causal association network for deviations by traversing a preset clinical medication knowledge graph based on the medication deviation report, including: Extract the atypical drugs corresponding to the medication order deviation, the specific drugs and excess values ​​corresponding to the dosage abnormality deviation, and the cost overrun drug categories corresponding to the cost growth abnormality deviation from the medication deviation report. Using the aforementioned atypical drugs, specific drugs, and drug categories with excessive costs as query entities, entity retrieval is performed in the clinical drug knowledge graph to locate the node position of the query entity in the knowledge graph. Read all preset types of relationship paths starting from nodes from the clinical drug knowledge graph. The preset types of relationship paths include drug substitution relationship paths, drug incompatibility relationship paths, treatment guideline recommendation relationship paths, and drug cost-effectiveness relationship paths. Perform a multi-hop traversal along each read relationship path, collect all associated intermediate entity nodes and relationship edges on the path, and form multiple subgraphs with the query entity as the root node. All collected subgraphs are merged, duplicate nodes and edges are removed, and a network is constructed that connects the various deviations in the medication deviation report with potential clinical and cost factors to establish the causes of the deviations.

5. The DRG-based clinical medication data management system according to claim 4, characterized in that, The strategy generation module retrieves a set of medication adjustment suggestions from a preset medication optimization rule base based on the deviation cause association network, including: The deviation cause correlation network is structurally decomposed into several independent deviation factor graphs. Each deviation factor graph describes a major deviation type and its associated causal path. For each deviation factor graph, extract its core deviation type and key causal entity, and combine the deviation type and key causal entity into a query statement. The query statement is used to perform pattern matching in the medication optimization rule base, which stores rules in the format of "deviation condition-cause condition-adjustment suggestion"; When the deviation type and key causal entity in the query statement match the deviation condition and causal condition in a rule, the rule is triggered and the corresponding adjustment suggestion is obtained. The adjustment suggestions obtained from all deviation factor plots will be summarized and deduplicated to form a set of medication adjustment suggestions for the new case.

6. The DRG-based clinical medication data management system according to claim 5, characterized in that, The strategy generation module sends the set of medication adjustment suggestions to the clinical decision support terminal, including: Each medication adjustment suggestion in the set of suggestions is prioritized, and the prioritization is calculated based on the severity of the impact of the deviation associated with the adjustment suggestion on treatment safety and cost. For each adjustment suggestion, a corresponding execution context description is generated, which includes the applicable treatment stage, the drugs involved, a summary of the expected adjustment effect, and reference basis; The sorted medication adjustment recommendations and their execution context descriptions are encapsulated into a decision support message in a specific format; The decision support message is pushed to the clinical decision support terminal of the target clinician through the medical data exchange interface and highlighted in a designated area of ​​the terminal interface.

7. The DRG-based clinical medication data management system according to claim 2, characterized in that, Within each treatment phase time window, the encoded drug usage sequence undergoes frequent pattern mining to identify high-frequency co-occurring drug combination patterns within the treatment phase time window, which are then used as typical drug combinations for the treatment phase. These include: Obtain coded drug use sequences for all cases within the current treatment phase time window, with each sequence containing multiple standard drug classification codes sorted by drug use time; Set a minimum support threshold, which represents the minimum frequency at which the drug combination appears in all case sequences; A frequent pattern growth algorithm is used to traverse all drug usage sequences, count the frequency of each drug code, and generate a header table by sorting them in descending order of frequency. Based on the head table, the drug use sequence of each case is projected to construct a conditional pattern base; Recursively mine frequent itemsets on the conditional pattern base. When the support of an itemset reaches the minimum support threshold, the drug combination is marked as a high-frequency combination. All identified high-frequency drug combinations are sorted in descending order of support, and the drug combinations with the highest support are selected as typical drug combinations for the current treatment phase time window. Typical drug combinations are bound and stored with corresponding treatment phase time windows for subsequent construction of a directed graph of medication patterns.

8. The DRG-based clinical medication data management system according to claim 3, characterized in that, The step of aligning and matching the medication timeline of the new case with the treatment stage time windows in the typical medication pattern atlas to determine the current treatment stage of the new case includes: Analyze the medication timeline of new cases, extracting the start time of medication, the time of each drug's use, and the current time. Predefined treatment phase time window division rules are obtained from typical medication pattern diagrams, the rules including the duration range and sequential relationship of each treatment phase; The starting time of medication for new cases is taken as the zero point of the time axis, and the offset of each medication usage time point relative to the zero point is calculated. Based on the duration range of the treatment phase time window, multiple consecutive treatment phase time windows are divided on the time axis. The drug use time sequence of new cases was matched with the time windows of the defined treatment stages, and the number of drug use events occurring in each time window was counted. The treatment phase with the highest number of drug use events was selected as the most likely current treatment phase for new cases. When there are multiple time windows with a similar number of drug use events, the time window with the highest time density is selected as the current treatment stage, taking into account the time density distribution characteristics of drug use. The final identified treatment stage identifier is output for subsequent drug use sequence alignment analysis.

9. The DRG-based clinical medication data management system according to claim 1, characterized in that, The system also includes a medication pattern update module, used to dynamically update the typical medication pattern map based on feedback data from new cases, including: Receive feedback data from the clinical decision support terminal, the feedback data recording the medication regimens actually adopted by doctors for new cases and the implementation status of the set of medication adjustment suggestions; Extract proven effective medication pattern fragments from adopted medication regimens; The similarity between the medication pattern fragment and the corresponding treatment stage node in the typical medication pattern atlas is calculated. If the similarity is higher than a preset threshold, the weight of the node in the atlas is strengthened. If the similarity is lower than the preset threshold but the medication result is well evaluated, the medication pattern fragment is added to the atlas as a new candidate pattern branch. Based on feedback data from all new cases within a time period, the weights and branch structures of each node in the typical medication pattern atlas are recalculated and optimized to generate an updated version of the typical medication pattern atlas.

10. The DRG-based clinical medication data management system according to claim 9, characterized in that, The medication pattern update module adds medication pattern fragments as new candidate pattern branches to the atlas, including: Analyze the differences between the medication pattern segment and existing atlas nodes in terms of drug composition, medication sequence and dosage, and locate the target node in the typical medication pattern atlas that is the same as the treatment stage of the medication pattern segment; Using the target node as the parent node, create new child nodes to represent the medication mode segment; Establish a directed edge between the parent node and the new child node, and label the specific differences between the parent node and the new child node on the directed edge; Initialize weights for new child nodes, which are calculated based on the number of cases that generated this medication pattern segment and the medication outcome assessment score.

Images