CN-122000089-A - Medicine risk analysis method and system based on full life cycle data

CN122000089ACN 122000089 ACN122000089 ACN 122000089ACN-122000089-A

Abstract

The invention discloses a medicine risk analysis method and a system based on full life cycle data, which belong to the technical field of medicine quality and safety management, and the method comprises the steps of constructing a medicine full life cycle causal network comprising nodes and directed edges based on medicine history and real-time full life cycle data; calculating node risk entropy of each node, performing risk conduction simulation to determine risk accumulation conditions, and predicting and outputting risk burst paths. The method is characterized in that self-adaptive evolution is realized through a double closed loop feedback mechanism, wherein a first closed loop corrects the causal network according to the comparison result of a predicted risk burst path and an actual risk event, and a second closed loop updates an ideal state baseline model according to global risk state analysis. The invention solves the technical problems of static state and poor adaptability of the existing risk analysis model, realizes the risk analysis capable of self-correction and evolution, and remarkably improves the long-term accuracy of risk prediction.

Inventors

ZHU TINGYU
HE MIAO
XIA MENGXUE
YI WENXUAN
TANG XIANGBIN
HUANG JIE
LI YUJIE
ZHOU YAN

Assignees

太极集团四川绵阳制药有限公司

Dates

Publication Date: 20260508
Application Date: 20260106

Claims (10)

1. The medicine risk analysis method based on the full life cycle data is characterized by comprising the following steps of: a. Constructing a medicine full life cycle causal network comprising a plurality of nodes and directed edges based on the history and real-time full life cycle data of the medicine, wherein the nodes represent entities or states in the medicine life cycle, and the directed edges represent causal relations among the nodes and are configured with conduction coefficients for the directed edges; b. Aiming at each node in the medicine full life cycle causal network, calculating to obtain node risk entropy of the node based on the deviation degree of the current state probability distribution of the node and a preset ideal state baseline model; c. performing risk conduction simulation based on the node risk entropy and the conduction coefficient of the directed edge of each node to determine the accumulation condition of risks in the drug whole life cycle causal network; d. predicting and outputting a risk burst path according to the accumulation condition of the risks; e. and acquiring a prediction result of the risk burst path and a verification result of an actually-occurring risk event, and dynamically correcting the medicine full life cycle causal network according to the verification result.
2. The method for analyzing risk of pharmaceutical product based on full life cycle data according to claim 1, wherein in the step a, The historical and real-time full life cycle data of the drug include: Data of four links of research and development design, production execution, circulation tracing and use feedback are covered; and timestamp data, temperature chain data and tracking code data throughout the four links; the construction of the medicine full life cycle causal network specifically comprises the following steps: And determining directed edges among the nodes by adopting a hybrid driving method combining a mechanism model and a data-driven causal inference algorithm.
3. The method for analyzing risk of pharmaceutical product based on full life cycle data according to claim 1, wherein in the step b, The node risk entropy of the node is calculated specifically including: And calculating the KL divergence of the current state probability distribution relative to the ideal state baseline model, and combining the KL divergence with a preset severity coefficient representing the inherent risk level of the node to obtain the node risk entropy.
4. The method for analyzing risk of pharmaceutical product based on full life cycle data according to claim 1, wherein in the step c, The risk conduction simulation is carried out specifically by: And calculating the input increment of the risk to the upstream node based on the node risk entropy of the upstream node, the conduction coefficient of the directed edge between the upstream node and the downstream node and a susceptibility coefficient representing the sensitivity degree of the downstream node to the upstream input risk.
5. The method for analyzing risk of pharmaceutical product based on full life cycle data according to claim 1, wherein in the step d, Predicting and outputting a risk burst path specifically includes: converting the medicine full life cycle causal network into a weighted graph taking the reciprocal of the conductivity coefficient as the passing cost; and searching and determining a minimum accumulated passing cost path from the risk source node to the final hazard node in the weighted graph by adopting a graph searching algorithm as the risk burst path.
6. The method for analyzing risk of pharmaceutical product based on full life cycle data according to claim 1, wherein in said step e, The dynamic correction of the drug full life cycle causal network specifically comprises: Comparing the predicted result of the risk burst path with an actual risk path corresponding to the actually-occurring risk event; and according to the comparison result, executing at least one of the following correction operations: Updating the conduction coefficients of the corresponding directed edges in the drug full life cycle causal network, wherein if the directed edges in the predicted path exist in the actual risk path, the conduction coefficients of the directed edges are increased; Adding new directed edges to the drug full life cycle causal network.
7. A method of drug risk analysis based on full lifecycle data as described in claim 1, the drug risk analysis method based on the full life cycle data is characterized by further comprising the following steps: Periodically analyzing node risk entropy of all nodes in the medicine full life cycle causal network, and identifying a stable domain in a low risk entropy state for a long term; updating the ideal state baseline model based on the data of the nodes in the stable domain.
8. A method of drug risk analysis based on full lifecycle data as described in claim 7, the drug risk analysis method based on the full life cycle data is characterized by further comprising the following steps: When the ideal state baseline model for a particular node is relaxed for that particular node, the conductivity coefficients of all directed edges pointed out by that particular node are compensatively increased.
9. The full life cycle data based drug risk analysis method of claim 1, wherein the full life cycle data based drug risk analysis method achieves adaptive evolution through a dual closed loop feedback mechanism comprising: A first closed loop feedback for correcting the directed edges and the conduction coefficients of the drug full life cycle causal network based on the comparison result of the risk burst path and the actual risk path; And a second closed loop feedback for adjusting the ideal state baseline model based on a global analysis of node risk entropy of all nodes in the drug full lifecycle causal network.
10. The drug risk analysis system (100) based on the full life cycle data is characterized by comprising a data acquisition module (101), a network construction module (102), a storage module (106), a risk calculation and transmission module (103), a path prediction module (104) and a dynamic correction module (105) which are communicated in sequence, and a processor (107) which is communicated with the data acquisition module (101), the network construction module (102), the risk calculation and transmission module (103), the path prediction module (104) and the dynamic correction module (105) respectively.

Description

Medicine risk analysis method and system based on full life cycle data Technical Field The invention relates to the technical field of medicine quality and safety management, in particular to a medicine risk analysis method and system based on full life cycle data. Background In the pharmaceutical industry, systematic risk management of the whole life cycle of products is a core link for guaranteeing the quality, safety and effectiveness of medicines, and is also a mandatory requirement of global medicine regulatory authorities (such as the FDA of the american food and medicine administration, EMA of the european medicine administration). Guidelines such as ICHQ < quality risk management > clearly advocate the use of prospective or retrospective methods to identify, evaluate, control, communicate and audit potential risks throughout the chain of product development, production, distribution until end use. Currently, risk assessment tools commonly used in industry, such as failure mode and impact analysis (FMEA), fault Tree Analysis (FTA), hazard analysis and key control point (HACCP), play an important role in primary identification and qualitative assessment of risks. These traditional methods typically rely on expert team knowledge and experience at a particular stage of the project (e.g., process development or technology transfer stage) to build a risk model through discussion and scoring. However, these existing technologies are increasingly limited in the face of increasingly complex and data-oriented modern pharmaceutical environments. One significant drawback is the static nature of existing risk models. Once the risk assessment is completed, its results, such as Risk Priority (RPN), remain unchanged for a long period of time, lacking dynamic association with the massive real-time data continuously generated during production. This results in gradual disconnection of the risk assessment from the actual operating conditions, and the model cannot automatically reflect risk baseline drift due to raw material changes, equipment aging, or process trimming. In addition, the existing methods have insufficient adaptability. When the production process is optimized, the system enters a new state that is more stable, or when an unexpected risk event actually occurs, the static model cannot realize self-learning and evolution. Any revision to the model relies on manual, periodic, lagging review conferences, which not only consume a lot of resources, but are slow to react to, and it is difficult to capture early signals of risk evolution. Therefore, these models often become an archive type compliance file, rather than a dynamic management tool capable of guiding daily operations and continuous evolution. Still further, the prior art also appears to be frustrating in quantifying the conductive and cumulative effects of risk. In a highly correlated production flow, minor fluctuations of an upstream parameter may cause significant quality deviations downstream through complex causal links. It is difficult for conventional methods to accurately simulate the propagation path and pooling effects of such risks, so that the risk conduction path most likely to explode and cause final hazard in the current system state cannot be accurately predicted. Therefore, how to overcome the static, hysteresis and subjectivity of the existing risk analysis method, develop a brand new technical scheme capable of deeply fusing full life cycle data, realizing dynamic correction and self-adaptive evolution of a risk model and accurately predicting a risk outburst path, and become a technical problem to be solved in the art. Disclosure of Invention Aiming at the defects of the prior art, one of the purposes of the invention is to provide a drug risk analysis method based on full life cycle data, so as to solve the problems that the existing risk analysis model is static and poor in adaptability, cannot carry out self-correction based on real-time data and actual risk events, and is difficult to quantitatively predict the conduction path and the cumulative effect of risks in a complex system. In order to achieve the above purpose, the invention is realized by the following technical scheme: A drug risk analysis method based on full life cycle data comprises the following steps: a. Constructing a medicine full life cycle causal network comprising a plurality of nodes and directed edges based on the history and real-time full life cycle data of the medicine, wherein the nodes represent entities or states in the medicine life cycle, and the directed edges represent causal relations among the nodes and are configured with conduction coefficients for the directed edges; b. Aiming at each node in the medicine full life cycle causal network, calculating to obtain node risk entropy of the node based on the deviation degree of the current state probability distribution of the node and a preset ideal state baseline model; c. performing risk condu