Drug use management system for hematology department

Abstract

The invention discloses a medicine use management system for the hematology department. According to the system, heterogeneous data of patients are fused through a multi-source data fusion module based on a Dempster-Shafer evidence theory algorithm, key genes are identified by a gene network construction module by adopting a PageRank algorithm introducing clinical correlation correction, and accurate matching is realized through a drug target matching module based on a collaborative filtering algorithm. The drug resistance prediction module adopts a time sequence decomposition method to realize dynamic prediction, the treatment scheme optimization module realizes multi-target balance optimization based on multi-criterion decision analysis, and the intelligent decision support module provides decision support based on an expert rule base. The technical problems that in the prior art, gene feature analysis is single, drug resistance prediction lacks dynamic modeling, and treatment scheme optimization targets are limited are solved, and individualized precise drug administration management on patients with recurrent / refractory hematologic tumors is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical information management, in particular to a drug use management system for hematology department. BACKGROUND

[0002] In the treatment of malignant hematological tumors such as relapsed / refractory leukemia and lymphoma, patients often have poor prognosis-related genes or secondary drug resistance genes, resulting in limited effect of traditional chemotherapy regimens. At present, there are target drugs and immunotherapy drugs such as PD-1 inhibitors targeting specific targets in clinical practice, but there are the following technical problems in drug selection and treatment regimen development process:

[0003] 1. The existing drug selection system mainly performs static matching based on single gene detection results, and fails to comprehensively consider the complex interaction relationship between primary drug resistance genes, secondary drug resistance genes and drug target genes, resulting in insufficient accuracy of drug sensitivity evaluation.

[0004] 2. The traditional drug resistance analysis method lacks quantitative modeling of the dynamic evolution process of drug resistance, and cannot predict the development trend of drug resistance during treatment, affecting the prospective adjustment of treatment regimen.

[0005] 3. The existing system usually adopts an empirical single-target optimization method in treatment regimen development, and fails to establish an effective multi-objective balancing mechanism among efficacy expectation, toxicity risk and drug resistance risk, resulting in limited optimization degree of individualized treatment regimen.

[0006] 4. The fusion processing technology of multi-source heterogeneous data such as gene detection data, clinical test data and imaging data is not mature, and the conflict and inconsistency between data affect the reliability of decision support.

[0007] In the prior art, although there are drug recommendation systems based on gene detection, these systems generally have the technical defects of single algorithm, insufficient data fusion capability, lack of dynamic prediction mechanism, etc., and cannot meet the technical needs of individualized precise treatment of complex hematological tumors.

[0008] Therefore, a drug use management system for hematology department is needed, which can comprehensively process multi-dimensional gene characteristics, realize dynamic prediction of drug resistance, and support multi-objective treatment regimen optimization. SUMMARY

[0009] The purpose of the present application is to provide a drug use management system for hematology department to solve the technical problems of single gene feature analysis, lack of dynamic modeling of drug resistance prediction, and limited optimization target of treatment regimen in the prior art.

[0010] To achieve the above purpose, the present application provides a drug use management system for hematology department, comprising:

[0011] The multi-source data fusion module is used to collect patients' gene testing data, clinical test data, and physiological indicator data, and to resolve conflicts and inconsistencies between different data sources. The multi-source data fusion module is based on the Dempster-Shafer evidence theory algorithm. By introducing data source weight coefficients and conflict detection mechanisms, it weights and fuses the basic probability allocations of multiple data sources to generate a unified patient feature dataset.

[0012] The gene network construction module receives the patient feature dataset output by the multi-source data fusion module, constructs a gene network graph based on the interaction relationship between genes, and adopts the PageRank algorithm and introduces gene connection weights and degree correction factors. By iteratively calculating the importance score of each gene node, it identifies primary drug resistance genes, secondary drug resistance genes and drug target genes that have a key impact on drug sensitivity.

[0013] The drug target matching module receives gene importance score data output by the gene network construction module, and calculates the matching probability between the drug and the target by analyzing the similarity relationship between drug molecular features and target gene features. The drug target matching module is based on a collaborative filtering algorithm, and uses the cosine similarity between drug feature vectors and pharmacological property vectors, combined with the distance metric of Gaussian kernel function, to generate a drug-target matching matrix.

[0014] The drug resistance prediction module receives the matching matrix data output by the drug target matching module and models and analyzes the development trajectory of the patient's drug resistance to a specific drug. The drug resistance prediction module uses the time series decomposition method to decompose the patient's historical drug resistance data into three components: long-term trend, periodic fluctuation and random disturbance. It predicts the future drug resistance index through trend extrapolation and dynamic weighting.

[0015] The treatment plan optimization module integrates the prediction results of the drug resistance prediction module and the drug target matching results to generate individualized treatment plans under the multi-objective constraints of efficacy expectation, toxic side effect risk, and drug resistance risk. The treatment plan optimization module is based on a multi-criteria decision analysis method, constructs a decision matrix containing each candidate drug plan, calculates the relative closeness of each plan to the ideal solution through standardization and weight allocation, and outputs a set of optimized plans after ranking.

[0016] The intelligent decision support module receives the set of treatment plans output by the treatment plan optimization module, combines it with the clinical expert knowledge rule base, and performs feasibility verification and risk assessment on the optimized plans. The intelligent decision support module assigns decision support weights to each treatment plan through rule matching degree calculation and activation intensity analysis, and generates a decision support report containing the reasons for the plan recommendation and precautions.

[0017] The hematology drug use management system of the present invention forms a complete technical process from data collection, feature analysis, matching calculation, risk prediction, scheme optimization to decision support through the coordinated work of the above six modules, realizing individualized and precise drug management for patients with relapsed / refractory hematological malignancies.

[0018] Furthermore, the Dempster-Shafer evidence theory-based algorithm used in the multi-source data fusion module improves the accuracy of heterogeneous data fusion by assigning credibility weights to each data source and establishing a conflict detection mechanism to identify contradictory information between data sources.

[0019] Furthermore, the gene network construction module introduces gene connection weights based on the PageRank algorithm. These weights reflect the differences in the strength of interactions between different genes. At the same time, a degree correction factor is added to balance the importance assessment bias between highly connected genes and poorly connected genes.

[0020] Furthermore, the drug target matching module performs dual similarity calculations by combining the molecular structure features and pharmacological properties of the drug. It measures the directional consistency between feature vectors through cosine similarity and measures the proximity of pharmacological properties through Gaussian kernel function.

[0021] Furthermore, the drug resistance prediction module decomposes the time series data into trend components, seasonal components, and residual components. The trend component reflects the long-term development direction of drug resistance, the seasonal component reflects the influence pattern of periodic treatment on drug resistance, and the residual component reflects the interference of random factors.

[0022] Furthermore, the multi-criteria decision analysis method adopted by the treatment plan optimization module includes four steps: decision matrix standardization, weight coefficient determination, ideal solution construction, and relative proximity calculation, which realizes the quantitative balance among multiple conflicting objectives.

[0023] Furthermore, the expert knowledge rule base of the intelligent decision support module includes drug contraindication rules, drug interaction rules, dosage adjustment rules, and monitoring requirement rules, and realizes individualized adaptation of decision support through dynamic adjustment of rule activation intensity.

[0024] Furthermore, the hematology drug use management system of the present invention, through the collaborative work of the above six modules, forms a complete technical process from data collection, feature analysis, matching calculation, risk prediction, scheme optimization to decision support, realizing individualized and precise drug management for patients with relapsed / refractory hematological malignancies.

[0025] Furthermore, the workflow of the hematology drug use management system includes:

[0026] S1. Multi-source data fusion: The multi-source data fusion module collects patients' gene testing reports, blood biochemical indicators, imaging examination results and clinical medical history information, and fuses heterogeneous data into a standardized patient feature dataset through conflict detection and weight allocation mechanisms;

[0027] S2. Gene Network Construction: The gene network construction module constructs a network map containing gene interaction relationships based on gene information in the fused data, calculates the importance score of each gene node, and identifies key genes that affect drug sensitivity.

[0028] S3. Drug Target Matching: The drug target matching module uses gene importance score results and molecular feature information in the drug database to calculate the matching degree between the patient's gene features and each candidate drug target, and generates a drug-target matching matrix.

[0029] S4. Drug Resistance Prediction: The drug resistance prediction module models and predicts the drug resistance development trend of each candidate drug based on the patient's historical medication records and current genetic characteristics, and outputs time series drug resistance risk assessment results.

[0030] S5. Treatment Plan Optimization: The treatment plan optimization module integrates drug matching degree and drug resistance risk prediction results, and generates a ranked set of individualized treatment plan candidates under the multi-objective constraints of efficacy expectation, toxic side effects and drug resistance risk.

[0031] S6. Intelligent Decision Support: The intelligent decision support module verifies the clinical feasibility of candidate treatment plans, and, in conjunction with an expert knowledge rule base, provides detailed decision support information and clinical precautions for each plan, generating a final treatment recommendation report.

[0032] Throughout the system's workflow, information is transmitted between modules through standardized data interfaces, ensuring the continuity and consistency of data processing. At the same time, it supports real-time data updates and dynamic plan adjustments, meeting the technical requirements for rapid response in clinical environments.

[0033] The beneficial effects of this invention are:

[0034] 1. Improved accuracy of multi-source data fusion: The Dempster-Shafer evidence theory-based algorithm, which introduces a conflict detection mechanism, can effectively identify and process contradictory information between different data sources, avoid the information distortion problem caused by simple data merging, and improve the accuracy and completeness of patient feature data.

[0035] 2. Enhanced accuracy of gene importance assessment: By introducing gene connection weights and degree correction factors into the PageRank algorithm, the true influence of genes in the interaction network can be reflected more accurately, avoiding the bias problem of excessive weights for highly connected genes in traditional methods.

[0036] 3. Improved reliability of drug target matching: By combining the dual similarity calculation of molecular structure features and pharmacological property features, a more comprehensive drug-target matching assessment is provided, overcoming the limitations of single feature matching and improving the reliability of matching results.

[0037] 4. Prospective realization of drug resistance prediction: Through dynamic modeling methods based on time series decomposition, it is possible to distinguish the different impacts of long-term trends and short-term fluctuations on drug resistance, thereby realizing quantitative prediction of future drug resistance development and providing forward-looking guidance for treatment regimen adjustments.

[0038] 5. Ensuring the balance of optimized treatment plans: Through quantitative methods of multi-criteria decision analysis, a scientific balance mechanism is established among multiple objectives such as efficacy, safety, and drug resistance risk, avoiding the subjectivity and one-sidedness of traditional empirical decision-making.

[0039] 6. Enhanced intelligence in decision support: By combining expert knowledge rule base and rule activation mechanism, the digital expression and intelligent application of clinical experience are realized, improving the professionalism and individualization of decision support. Attached Figure Description

[0040] Figure 1 The system architecture flowchart of the present invention is shown;

[0041] Figure 2 A flowchart of the steps of the system described in this invention is shown. Detailed Implementation

[0042] The hematology drug use management system of the present invention will now be described in detail with reference to the accompanying drawings.

[0043] like Figure 1 As shown, the hematology drug use management system of the present invention adopts a modular architecture design and runs in a distributed computing environment. The hardware environment of the hematology drug use management system includes: a server cluster configured with Intel Xeon E5-2680 processors (2.7GHz, 8 cores), 64GB DDR4 memory, and 2TB SSD solid-state drive storage, connected via gigabit Ethernet, supporting concurrent processing under the Linux operating system environment.

[0044] Furthermore, the software architecture of the hematology drug use management system consists of six core functional modules, the functions of which are as follows:

[0045] First, the multi-source data fusion module, located in the data input layer of the system architecture, is responsible for acquiring heterogeneous patient data from the hospital information system, laboratory information management system, and gene testing equipment interface. This module acquires data via a standardized HL7 FHIR protocol interface, supporting both real-time data streaming and batch data import modes. Specifically, it acquires gene testing data (supporting VCF and MAF formats), clinical laboratory data (including complete blood count, biochemical indicators, and coagulation function test results), physiological indicator data (vital signs monitoring data, imaging examination results, and pathology reports), and clinical history information (past medical history records, medication records, and treatment response data) through standardized interfaces. The module performs standardized processing on the acquired heterogeneous data, including converting the data format to the system's internal standard JSON format, unifying units to international standard units, adopting ICD-10 disease coding and SNOMED CT clinical terminology standard coding systems, and using multiple imputation methods to handle missing data.

[0046] Based on the Dempster-Shafer evidence theory algorithm, the multi-source data fusion module achieves intelligent fusion of multi-source data. First, the multi-source data fusion module calculates the weight coefficient wi for each data source i:

[0047]

[0048] Where Qi is the data quality score of data source i, which is calculated based on data integrity, accuracy and consistency, and its value range is [0,1]; Ti is the timeliness coefficient of data source i, which reflects the freshness of the data, and is calculated as Ti=exp(-λ·Δt), where λ is the timeliness decay parameter and Δt is the time difference from the time the data was generated to the current time; Ci is the credibility coefficient of data source i, which is determined based on the authority of the data source, with 0.9 for hospital information systems, 0.95 for laboratory information systems and 0.85 for gene testing equipment; and n is the total number of data sources participating in the fusion.

[0049] Next, the multi-source data fusion module calculates the basic probability allocation function:

[0050]

[0051] Where mi(A) is the basic probability assignment of data source i to hypothesis A, with a value range of [0,1]; Qi(A) is the quality assessment value of data source i to hypothesis A, calculated based on the degree of support of the data source for a specific hypothesis; Ri is the overall reliability coefficient of data source i, which comprehensively considers the historical accuracy and technological maturity of the data source; Θ is the complete hypothesis space, which contains all possible patient disease state hypotheses; and B is any subset of the hypothesis space.

[0052] Then, the multi-source data fusion module calculates the evidence conflict coefficient:

[0053]

[0054] Here, K represents the degree of conflict between evidence from different data sources. When K is close to 0, it indicates high consistency of evidence, and when K is close to 1, it indicates high conflict of evidence.

[0055] Subsequently, the multi-source data fusion module performs evidence fusion:

[0056]

[0057] Where m(A) is the basic probability assignment of hypothesis A after fusion.

[0058] Finally, the multi-source data fusion module updates the confidence level:

[0059]

[0060]

[0061] Where Bel(A) is the confidence function value of hypothesis A, representing the minimum confidence level of hypothesis A; Pl(A) is the likelihood function value of hypothesis A, representing the maximum possible confidence level of hypothesis A; Let be the complement of hypothesis A in hypothesis space Θ.

[0062] Next, the gene network construction module receives standardized data output from the multi-source data fusion module and constructs a gene interaction network based on the Neo4j graph database. This module maintains a network map containing over 20,000 protein-coding genes in the human genome, supporting dynamic updates and incremental construction. Specifically, the module constructs a comprehensive gene interaction knowledge base based on multiple authoritative databases, including protein-protein interaction networks from the STRING database, gene functional pathway information from the KEGG database, biological response pathway data from ReactomeDB, and gene-disease association information from the OMIM database.

[0063] The gene network construction module uses an adjacency matrix W to represent the gene interaction network:

[0064]

[0065] Where wij represents the interaction strength between gene i and gene j, with a value range of [0,1], and is calculated using the following formula:

[0066]

[0067] Where Sij represents the sequence similarity between genes i and j, calculated using the BLAST alignment algorithm; Pij represents the pathway co-occurrence frequency of genes i and j, indicating the frequency of the two genes appearing in the same biological pathway; Cij represents the co-expression correlation between genes i and j, calculated based on gene expression data using the Pearson correlation coefficient; α, β, and γ are weighting coefficients, set to 0.4, 0.3, and 0.3 respectively, satisfying α+β+γ=1.

[0068] Based on the standard PageRank algorithm, the gene network construction module incorporates clinical relevance correction:

[0069]

[0070] Wherein, PR(Gi) is the PageRank importance score of gene Gi, with a value range of [0,1]; d is the damping coefficient, representing the probability of continuing along the link during a random walk, usually set to 0.85; N is the total number of gene nodes in the network; M(Gi) is the set of gene nodes pointing to gene Gi; wji is the connection weight from gene Gj to gene Gi; L(Gj) is the out-degree of gene Gj, i.e., the number of connections from gene Gj to other genes; and CF(Gi) is the clinical relevance correction factor for gene Gi.

[0071] Furthermore, the gene network construction module calculates a clinical relevance correction factor:

[0072]

[0073] Wherein, Fdisease(Gi) is the association frequency between gene Gi and hematologic malignancies, calculated by counting the number of times the gene appears in hematologic malignancies-related literature; Fdrug(Gi) is the frequency of gene Gi as a drug target, calculated by counting the number of times the gene is a known drug target; λ is a correction strength parameter used to control the degree of influence of clinical relevance on PageRank score, usually set to 0.1.

[0074] The convergence condition for the gene network construction module is set as follows:

[0075]

[0076] Where PR(t)(Gi) is the PageRank value of gene Gi at the t-th iteration; PR(t+1)(Gi) is the PageRank value of gene Gi at the (t+1)-th iteration; ε is the convergence threshold, set to 10⁻⁶; and t is the number of iterations.

[0077] Finally, the gene network construction module calculates the gene influence score:

[0078]

[0079] Where IS(Gi) is the final influence score of gene Gi; deg(Gi) is the degree of gene Gi in the network, that is, the number of genes directly connected to gene Gi.

[0080] Furthermore, the drug target matching module utilizes the ChEMBL drug database and the UniProt protein database to maintain feature data for over 10,000 drug molecules. The module calculates molecular descriptors using the RDKit cheminformatics toolkit, supporting similarity analysis of 2D and 3D molecular features.

[0081] Specifically, the drug target matching module extracts a multidimensional feature vector for each drug molecule. The molecular structure feature vector Fi includes a 1024-bit binary vector of the molecular fingerprint ECFP4, a molecular descriptor (including molecular weight MW, octanol-water partition coefficient LogP value, number of hydrogen bond donors HBD, number of hydrogen bond acceptors HBA), and a topological fingerprint based on the molecular graph. Meanwhile, the pharmacological property vector Pi includes absorption properties (oral bioavailability F, intestinal absorption rate Papp), distribution properties (blood-brain barrier permeability logBB, plasma protein binding rate PPB), metabolic properties (cytochrome P450 enzyme substrate characteristics CYP, cytochrome P450 enzyme inhibitor characteristics), and excretion properties (renal clearance CLr, drug half-life t1 / 2).

[0082] The drug target matching module uses multi-feature fusion to calculate drug-target similarity:

[0083]

[0084] Wherein, Sim(Di,Tj) is the comprehensive similarity score between drug Di and target Tj, with a value range of [0,1]; ω1 and ω2 are feature weights, set to 0.6 and 0.4 respectively, satisfying ω1+ω2=1; Simstruct(Di,Tj) is the structural similarity; and Simpharm(Di,Tj) is the pharmacological similarity.

[0085] The drug target matching module calculates structural similarity:

[0086]

[0087] Where Fi·Fj is the dot product of eigenvectors Fi and Fj; ||Fi|| is the Euclidean norm of eigenvector Fi.

[0088] Simultaneously, the drug target matching module calculates pharmacological similarity:

[0089]

[0090] Where, ‖Pi-Pj‖2 is the Euclidean distance between the pharmacological property vectors Pi and Pj; σ is the bandwidth parameter of the Gaussian kernel function, the optimal value is determined by cross-validation, and is usually set to 1.0.

[0091] Based on the collaborative filtering algorithm, the drug target matching module predicts the matching probability using the known potency of similar drugs:

[0092]

[0093] Wherein, P(Di,Tj) is the matching probability of drug Di to target Tj, with a value range of [0,1]; N(Di) are the K drug neighbors most similar to drug Di, and K is usually set to 10; R(Dk,Tj) is the known effect strength score of drug Dk on target Tj, which is derived from experimental validation data in the drug database.

[0094] Accordingly, the drug target matching module adjusts the confidence level:

[0095]

[0096] Where Conf(Di,Tj) is the adjusted confidence score; |N(Di)| is the number of neighbors of drug Di; and |N(Tj)| is the number of neighbors of target Tj.

[0097] Furthermore, the drug resistance prediction module stores patients' historical medication records and drug resistance development data based on the time-series database InfluxDB. The module employs a sliding window mechanism to process time-series data; the window size can be adaptively adjusted according to the data frequency, set to 30–90 days for daily frequency data and 12–24 weeks for weekly frequency data.

[0098] Specifically, the drug resistance prediction module preprocesses patients' historical medication and drug resistance data, including data cleaning (using the 3σ rule to remove outliers and noisy data), time alignment (aligning observation data of different frequencies to the same time grid), and interpolation completion (using cubic spline interpolation to fill in missing time point data).

[0099] The drug resistance prediction module uses a seasonal trend decomposition method to decompose the time series:

[0100]

[0101] Where X(t) is the observed value of drug resistance at time t; T(t) is the trend component, reflecting the long-term trend of drug resistance; S(t) is the seasonal component, reflecting the periodic fluctuations related to the treatment cycle; and R(t) is the residual component, reflecting random disturbances and unpredictable factors.

[0102] The drug resistance prediction module uses locally weighted regression to extract trend components.

[0103]

[0104] Where wi(t) is the weighting function, n is the total number of observation points, and ti is the i-th observation time point. The weighting function is calculated as follows:

[0105]

[0106] Where W(·) is the three-cubic kernel function, defined as W(u)=(1-|u|3)3 when |u|<1, otherwise W(u)=0; h is the bandwidth parameter, which controls the smoothness of the local regression.

[0107] Meanwhile, the drug resistance prediction module extracts a seasonal component:

[0108]

[0109] Where L is the seasonal cycle length, which is usually set to the treatment cycle length for hematologic malignancy treatment; ns is the number of seasonal cycles including time t; and j is the seasonal cycle index.

[0110] Furthermore, the drug resistance prediction module constructs a dynamic drug resistance prediction model:

[0111]

[0112] Wherein, DR(t+h) is the drug resistance index at the predicted time t+h; h is the prediction step size, representing the predicted drug resistance over the next h time units; Tpred(t+h) is the trend prediction value; Spred(t+h) is the seasonal prediction value; μR is the mean of the residual term; α, β, and γ are weighting parameters, estimated by the least squares method, typically α=0.6, β=0.3, and γ=0.1.

[0113] The drug resistance prediction module uses adaptive linear extrapolation for trend prediction:

[0114]

[0115] Where T(t) is the trend value at the current moment; ΔT(t) is the first difference of the trend, ΔT(t) = T(t) - T(t-1); Δ2T(t) is the second difference of the trend, Δ2T(t) = ΔT(t) - ΔT(t-1).

[0116] Simultaneously, the drug resistance prediction module calculates the prediction confidence interval:

[0117]

[0118] Where CI(t+h) is the confidence interval of the predicted value; zα / 2 is the α / 2 quantile of the standard normal distribution, and for the 95% confidence interval, z0.025=1.96; This is for estimating the variance of the prediction error.

[0119] The drug resistance prediction module calculates the prediction error variance:

[0120]

[0121] in, The variance of the trend component; The variance of the seasonal component; denoted as the variance of the residual components.

[0122] Furthermore, the treatment plan optimization module integrates multiple decision analysis algorithm libraries, including the scikit-learn machine learning library and the CVXPY optimization library. The treatment plan optimization module supports solving multi-objective optimization problems and can handle both linear and nonlinear constraints.

[0123] Specifically, the treatment plan optimization module models the treatment plan selection problem as a multi-objective optimization problem:

[0124]

[0125] Where F(x) is the objective function vector; x is the decision variable vector, representing the parameter configuration of the treatment plan. The objective function is defined as follows: f1(x)=-E(x) is the expected value of negative efficacy (maximizing efficacy by minimizing negative values); f2(x)=T(x) is the risk value of toxic side effects (minimizing the risk of toxicity); f3(x)=R(x) is the risk value of drug resistance (minimizing the risk of drug resistance development).

[0126] The treatment plan optimization module sets constraints:

[0127]

[0128]

[0129] Where gi(x) is the i-th inequality constraint, including upper and lower limits of drug dosage, dosing frequency limit, and contraindication constraint; hj(x) is the j-th equality constraint, including drug compatibility requirements and dosage balance constraint; m is the number of inequality constraints; and p is the number of equality constraints.

[0130] The treatment plan optimization module uses an approximation-ideal-solution ranking method to evaluate the plans, first constructing a standardized decision matrix:

[0131]

[0132] Where rij is the standardized score; xij is the original score of the i-th scheme under the j-th criterion; and m is the total number of candidate schemes.

[0133] Then, the treatment plan optimization module performs weighted standardization:

[0134]

[0135] Where vij is the weighted standardized score; wj is the weight of the j-th criterion, determined by the analytic hierarchy process, satisfying the following conditions: n represents the number of evaluation criteria.

[0136] Next, the treatment plan optimization module determines the positive ideal solution A+ and the negative ideal solution A-:

[0137]

[0138]

[0139] Among them, for the benefit-based criterion, , For cost-based criteria, , .

[0140] The treatment plan optimization module calculates the Euclidean distance from each plan to the ideal solution:

[0141]

[0142]

[0143] in, Let i be the distance from the ideal solution to the solution i. Let be the distance from scheme i to the negative ideal solution.

[0144] Finally, the treatment plan optimization module calculates the relative similarity:

[0145]

[0146] Where Ci is the relative proximity of scheme i, and its value ranges from [0,1]. The larger the value of Ci, the better the scheme.

[0147] Finally, the intelligent decision support module builds an expert knowledge system based on the Drools rule engine, maintaining over 1000 clinical decision rules. This module supports dynamic updates and version management, ensuring the timeliness and accuracy of clinical knowledge.

[0148] Specifically, the expert knowledge rule base of the intelligent decision support module adopts a production rule representation:

[0149]

[0150] Where Cki is the i-th antecedent condition of the k-th rule; Dk is the conclusion of the k-th rule; CFk is the confidence factor of the k-th rule, with a value range of [0,1]; p is the number of antecedent conditions; ∧ represents the logical AND operation.

[0151] The intelligent decision support module uses fuzzy matching to handle uncertainty information.

[0152]

[0153] Where μmatch(Cki,Fi) is the fuzzy matching degree function, with a value range of [0,1]; Cki is the expected value of the rule condition; Fi is the observed fact value; σi is the matching tolerance parameter, which controls the strictness of the matching.

[0154] Furthermore, the intelligent decision support module calculates the rule matching degree:

[0155]

[0156] Where Match(Rk,F) is the matching degree between rule Rk and fact F, with a value range of [0,1]; wki is the weight of the i-th condition in the k-th rule, satisfying... .

[0157] The intelligent decision support module calculates the activation strength of the rules:

[0158]

[0159] Where Act(Rk) is the activation strength of rule Rk; Impk is the importance weight of rule Rk, which is determined based on the clinical importance and frequency of use of the rule.

[0160] The intelligent decision support module adopts a conflict resolution strategy: high confidence priority (selecting the rule with the highest CFk), specificity priority (selecting the rule with the most specific antecedent conditions), and time proximity priority (selecting the most recently updated rule).

[0161] Finally, the intelligent decision support module calculates the decision weights:

[0162]

[0163] Where W(Dj) is the comprehensive weight of decision Dj; K is the total number of rules in the rule base; the summation condition "k: Dk=Dj" means that all rules have the conclusion Dj.

[0164] Combination Figure 2 The workflow of the hematology drug use management system includes:

[0165] S1. Multi-source data fusion: The multi-source data fusion module collects patients' gene testing reports, blood biochemical indicators, imaging examination results and clinical medical history information, and fuses heterogeneous data into a standardized patient feature dataset through conflict detection and weight allocation mechanisms;

[0166] S2. Gene Network Construction: The gene network construction module constructs a network map containing gene interaction relationships based on gene information in the fused data, calculates the importance score of each gene node, and identifies key genes that affect drug sensitivity.

[0167] S3. Drug Target Matching: The drug target matching module uses gene importance score results and molecular feature information in the drug database to calculate the matching degree between the patient's gene features and each candidate drug target, and generates a drug-target matching matrix.

[0168] S4. Drug Resistance Prediction: The drug resistance prediction module models and predicts the drug resistance development trend of each candidate drug based on the patient's historical medication records and current genetic characteristics, and outputs time series drug resistance risk assessment results.

[0169] S5. Treatment Plan Optimization: The treatment plan optimization module integrates drug matching degree and drug resistance risk prediction results, and generates a ranked set of individualized treatment plan candidates under the multi-objective constraints of efficacy expectation, toxic side effects and drug resistance risk.

[0170] S6. Intelligent Decision Support: The intelligent decision support module verifies the clinical feasibility of candidate treatment plans, and, in conjunction with an expert knowledge rule base, provides detailed decision support information and clinical precautions for each plan, generating a final treatment recommendation report.

[0171] In summary, this invention, through the aforementioned technologies, establishes a complete hematology drug use management system, realizing an end-to-end technical solution from multi-source data fusion to intelligent decision support. It effectively solves the technical problems of existing technologies, such as single gene feature analysis, lack of dynamic modeling for drug resistance prediction, and limited goals for treatment plan optimization, providing reliable technical support for individualized precision treatment of hematological malignancies.

[0172] While preferred embodiments of the invention have been described, those skilled in the art will understand that various changes, modifications, substitutions, and variations 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 hematology drug administration management system, characterized in that, include: The multi-source data fusion module is used to collect patients' gene testing data, clinical test data, and physiological indicator data. The multi-source data fusion module is based on the Dempster-Shafer evidence theory algorithm. By calculating the data source weight coefficient and evidence conflict coefficient, the basic probability allocation of multiple data sources is weighted and fused to generate a unified patient feature dataset. The gene network construction module receives the patient feature dataset output by the multi-source data fusion module, constructs a gene network graph based on the interaction relationship between genes, and adopts the PageRank algorithm and introduces a clinical relevance correction factor. By iteratively calculating the importance score of each gene node, it identifies key genes that affect drug sensitivity. The drug target matching module receives gene importance score data output by the gene network construction module, and generates a drug-target matching matrix based on the similarity calculation of drug feature vectors and pharmacological property vectors using a collaborative filtering algorithm. The drug resistance prediction module receives the matching matrix data output by the drug target matching module, and uses a time series decomposition method to decompose the patient's historical drug resistance data into trend components, seasonal components and residual components, and uses a dynamic prediction model to predict the future drug resistance index. The treatment plan optimization module integrates the prediction results of the drug resistance prediction module and the drug target matching results, and generates an individualized treatment plan based on a multi-criteria decision analysis method under the multi-objective constraints of efficacy expectation, toxic side effect risk, and drug resistance risk. The intelligent decision support module receives the set of treatment plans output by the treatment plan optimization module, and provides decision support weights and risk assessment information for the treatment plans based on the expert knowledge rule base, through rule matching degree calculation and activation intensity analysis.

2. The hematology drug administration management system according to claim 1, characterized in that, The multi-source data fusion module calculates weighted coefficients for each data source, including data quality score, timeliness coefficient, and credibility coefficient, and achieves consistent fusion of heterogeneous data based on an evidence conflict detection mechanism.

3. The hematology drug administration management system according to claim 1, characterized in that, The clinical relevance correction factor in the gene network construction module is calculated based on the association frequency between genes and hematological malignancies and the frequency of genes as drug targets, and is used to adjust the importance score of genes in the network.

4. The hematology drug administration management system according to claim 1, characterized in that, The drug target matching module achieves multi-dimensional feature similarity assessment between drugs and targets by calculating the cosine similarity of drug molecular structural features and the Gaussian kernel function distance of pharmacological property features.

5. The hematology drug administration management system according to claim 1, characterized in that, The drug resistance prediction module uses local weighted regression to extract trend components, implements trend prediction through adaptive linear extrapolation, and calculates the confidence interval of the prediction results.

6. The hematology drug administration management system according to claim 1, characterized in that, The treatment plan optimization module adopts the approximation ideal solution ranking method. By constructing a standardized decision matrix, determining positive and negative ideal solutions, and calculating Euclidean distance and relative proximity, it realizes the quantitative evaluation and ranking of treatment plans.

7. The hematology drug administration management system according to claim 1, characterized in that, The intelligent decision support module uses production rules to represent expert knowledge, processes uncertain information through fuzzy matching, and employs conflict resolution strategies that prioritize high confidence, specificity, and time proximity to achieve rule activation and decision weight allocation.

Images