Intelligent decision-making and traceability method for medical instrument procurement based on dynamic perception

Abstract

The application relates to the technical field of intelligent management of a medical instrument supply chain, and discloses a medical instrument procurement intelligent decision-making and tracing method based on dynamic sensing, which comprises the following steps: collecting current dynamic data; the current dynamic data comprises supplier basic data, public opinion data, supervision data and logistics data associated with the medical instrument in a current period; based on the current dynamic data, current time-varying covariates are generated; the current time-varying covariates are subjected to prediction processing through a random survival forest, and current predicted cumulative risks are obtained; the current predicted cumulative risks are associated with the medical instrument and / or a supplier of the medical instrument; the current predicted cumulative risks and the current time-varying covariates are subjected to dynamic processing through a conditional survival probability prediction function of a Cox model, so as to sense a current predicted risk probability associated with the medical instrument and / or the supplier; and a procurement decision for the medical instrument is determined based on the current predicted risk probability. The scheme can meet the actual procurement requirements of the medical instrument.

Description

Technical Field

[0001] This application relates to the field of intelligent management technology for medical device supply chains, specifically to, but not limited to, an intelligent decision-making and traceability method for medical device procurement based on dynamic perception. Background Technology

[0002] In the field of medical device procurement, dynamic changes in factors such as product quality and logistical conditions can impact the procurement process. To mitigate the negative impact of these factors, related technologies offer solutions relying on supplier qualification reviews and post-procurement responses. However, these solutions suffer from significant delays.

[0003] To address the aforementioned technical challenges, the technology also provides real-time tracking of the usage status of medical devices and the service capabilities of suppliers, and adjusts procurement decisions for medical devices based on the results of this real-time tracking.

[0004] However, the above solutions are based on adjusting procurement decisions according to the existing risk situation and cannot meet the actual procurement needs of medical devices. Summary of the Invention

[0005] Based on the above technical problems, this application provides a method for intelligent decision-making and traceability of medical device procurement based on dynamic perception, which can meet the actual procurement needs of medical devices.

[0006] The technical solution provided in this application is as follows: This application provides a method for intelligent decision-making and traceability in medical device procurement based on dynamic perception, including: Collect current dynamic data; wherein, the current dynamic data includes at least the basic data of the suppliers associated with the medical device, public opinion data, regulatory data, and logistics data for the current time period; Based on the current dynamic data, generate the current time-varying covariate; The current time-varying covariate is predicted using a random survival forest to obtain the current predicted cumulative risk; wherein the current predicted cumulative risk is associated with the medical device and / or the supplier of the medical device. The current predicted cumulative risk and the current time-varying covariate are dynamically processed using the conditional survival probability prediction function of the Cox model to perceive the current predicted risk probability associated with the medical device and / or the supplier. The procurement decision for the medical device is determined based on the current predicted risk probability; wherein the procurement decision includes risk tracing information for at least a portion of the current dynamic data.

[0007] The intelligent decision-making and traceability method for medical device procurement based on dynamic perception provided in this application has at least the following beneficial effects: The intelligent decision-making and traceability method for medical device procurement based on dynamic perception provided in this application collects current dynamic data, which includes basic supplier data, public opinion data, regulatory data, and logistics data related to the medical device within the current time period. Thus, through this current dynamic data, the actual status of the medical device within the supplier, public opinion, regulatory, and logistics dimensions can be comprehensively and diversely characterized within the current time period. Furthermore, based on the current dynamic data, a current time-varying covariate is generated. This current time-varying covariate more accurately reflects the patterns of data change over time. On this basis, a random survival forest is used to predict the current time-varying covariate, obtaining the current predicted cumulative risk. This current predicted cumulative risk is associated with the medical device and / or its supplier. Thus, by leveraging the random survival forest, the importance of features can be quantified and key risk factors can be identified, achieving… The advantages of dynamic, accurate, and interpretable risk prediction can improve the accuracy of current cumulative risk prediction. On the other hand, by using the conditional survival probability prediction function of the Cox model, the current cumulative risk prediction and the current time-varying covariate are dynamically processed to perceive the current predicted risk probability associated with medical devices and / or capture suppliers. Thus, by leveraging the dynamic risk tracking advantage of the Cox model, the accuracy of the current predicted risk probability can be improved. Furthermore, procurement decisions for medical devices are determined based on the current predicted risk probability, and the procurement decisions include risk tracing information from at least a portion of the current dynamic data. This not only achieves fine-grained risk tracing of at least a portion of the current dynamic data affecting procurement decisions, but also enables the pre-determination of procurement decisions for medical devices. This overcomes the shortcomings of lagging adjustments in related technology procurement decisions and meets the actual procurement needs for medical devices. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating the intelligent decision-making and traceability method for medical device procurement based on dynamic perception, provided in an embodiment of this application. Detailed Implementation

[0009] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0010] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0011] In the field of medical device procurement, dynamic changes in factors such as product quality and logistical conditions can impact the procurement process. To mitigate the negative impact of these factors, related technologies offer solutions relying on supplier qualification reviews and post-procurement responses. However, these solutions suffer from significant delays.

[0012] To address the aforementioned technical challenges, the technology also provides real-time tracking of the usage status of medical devices and the service capabilities of suppliers, and adjusts procurement decisions for medical devices based on the results of this real-time tracking.

[0013] However, the above solutions are based on adjusting procurement decisions according to the existing risk situation and cannot meet the actual procurement needs of medical devices.

[0014] To address the above technical issues, this application provides a method for intelligent decision-making and traceability in medical device procurement based on dynamic perception. Figure 1 This is a flowchart illustrating the intelligent decision-making and traceability method for medical device procurement based on dynamic perception provided in the embodiments of this application, as follows: Figure 1 As shown, the process may include the following steps: Step 101: Collect current dynamic data.

[0015] The current dynamic data includes at least the basic data of medical device suppliers, public opinion data, regulatory data, and logistics data for the current period.

[0016] In some embodiments, the current time period may include a period of time from a historical moment to the current moment; for example, the length of the current time period may be predetermined or adjusted, and this application embodiment does not limit this.

[0017] In some embodiments, a medical device may include instruments, equipment, appliances, in vitro diagnostic reagents and calibrators that can be used directly or indirectly on the human body and can function through physical, mechanical and electronic means.

[0018] In some embodiments, supplier basic data may include the supplier's main information, product and capability information, business and cooperation information, and medical device industry-specific fields. For example, product and capability information may include the supplier's main products, production capacity, and quality capabilities. For example, main information may include the supplier's name, unified social credit code, registered address, production address, date of establishment, registered capital, and enterprise type. For example, business and cooperation information may include cooperation status, supply level, delivery terms, price information, and contracts and quality agreements. For example, medical device industry-specific fields may include the sterility level or sterilization method of the medical device, scope of application, contact method, raw material compliance information, and traceability requirements.

[0019] Accordingly, supplier basic data can be obtained through any of the following methods: Uploaded voluntarily by the medical device supplier.

[0020] Obtained through the National Enterprise Credit Information Publicity System.

[0021] Obtained through the National Medical Products Administration (NMPA) database or the NMPA website.

[0022] In some embodiments, public opinion data may include news data related to medical devices and their suppliers that are publicly available in official media, self-media, short video platforms, relevant forums or industry platforms; for example, public opinion data may present medical device-related recall events, quality, sampling results, administrative penalties, bidding risks, qualification abnormalities and complaints from a positive, neutral or negative perspective.

[0023] Accordingly, public opinion data can be collected in the following ways: According to the collection strategy, based on target keywords, data carried in official media, self-media, short video platforms, related forums or industry platforms are searched and filtered to obtain public opinion data. For example, the collection strategy may include collection cycle or event-triggered collection. Event-triggered collection may include monitoring official media, self-media, short video platforms, related forums or industry platforms, etc., and if a target event related to medical devices is detected, the collection of public opinion data is triggered.

[0024] In some embodiments, regulatory data may be released by drug regulatory authorities; specifically, it may include dimensions such as supplier qualifications, registration, production, operation, sampling inspection, recall, adverse events, and UDI.

[0025] Accordingly, regulatory data can be obtained by filtering the data released by the National Medical Products Administration.

[0026] In some embodiments, logistics data may include end-to-end data on the warehousing, transportation, distribution, traceability, and temperature control of medical devices; correspondingly, logistics data can be collected in the following ways: Logistics data was obtained by filtering relevant data published on the official websites of medical device logistics companies.

[0027] In some embodiments, the current dynamic data can be represented in the form of a data sequence; specifically, the current dynamic data can be collected in the following ways: Define lifespan for each supplier and medical device If a risk event occurs during the observation period within the current time period, then The time of occurrence of the risk event; if no risk event occurs during the observation period, then... The observation end time or right censoring time is defined; a risk event indicator variable is also defined. If a risk event occurs, then set The value is 1; if no risk event occurs, it is set to 1. The value is 0; then, the data corresponding to T is integrated to obtain the unique identifier of the medical device supplier and the survival time. and risk event indicator variables The data sequence is then processed; next, the data sequences corresponding to each T within the current time period are integrated to obtain the current dynamic data. For example, the unique identifier of a supplier may include the supplier name and / or the unified social credit code.

[0028] For example, a risk event may include data on the degree of impact on the procurement process of medical devices that is greater than or equal to a threshold. Specifically, a risk event can be identified by integrating and analyzing supplier basic data, public opinion data, regulatory data, and logistics data. For instance, the supplier name and / or unified social credit code can be determined based on the supplier basic data. The analysis results can be obtained by analyzing at least one of the public opinion data, regulatory data, and logistics data based on the supplier name and / or unified social credit code. If the analysis results show that the sentiment of the supplier's public opinion data is negative, or the regulatory data indicates the occurrence of a recall or adverse event, or the logistics data indicates that the number of medical device logistics anomalies within a unit of time is greater than or equal to a frequency threshold, then a risk event related to the medical device and / or the supplier can be identified.

[0029] Specifically, NLP can be used to score the sentiment data contained in the data sequence. If the sentiment score is less than the first threshold, it can be determined that a risk event occurred within the time period corresponding to the data sequence. For example, if the regulatory data contained in the data sequence indicates that at least one of the following has occurred within a specified time period: at least one quality notice or surprise inspection has been carried out on the supplier, the quality pass rate is less than or equal to the second threshold, or the number of adverse events is greater than or equal to the third threshold, it can be determined that a risk event occurred within the time period corresponding to the data sequence. For another example, if the logistics data contained in the data sequence indicates that the logistics anomaly rate is greater than or equal to the fourth threshold, it can be determined that a risk event occurred within the time period corresponding to the data sequence.

[0030] Step 102: Generate the current time-varying covariates based on the current dynamic data.

[0031] In some embodiments, the current time-varying covariate may include a counting process format of the data sequence in the current dynamic data; for example, the current time-varying covariate may represent a multidimensional feature vector corresponding to the current dynamic data within a time window of the current period, or it may represent a snapshot of the risk status of the medical device and / or its supplier within the above time window.

[0032] Accordingly, the current time-varying covariates can be generated in the following way: The current time period is divided into segments according to a fixed time window to obtain a set of time segments. Then, segment data corresponding to the segment times in the time segment set is obtained from the current dynamic data to obtain a set of segment data. Dynamic features are extracted from the segment data in the set of segment data to obtain segment dynamic features. At this point, the segment dynamic features can be determined as time-varying covariates corresponding to the segment times. Then, the set of time-varying covariates corresponding to each segment time is determined as the current time-varying covariate.

[0033] For example, segmented dynamic features can be obtained in the following way: If the risk event indicator variable in the segmented data of the segmented dataset is defined as having a value of 1, then dynamic feature extraction can be performed on the segmented data in the segmented dataset to obtain time-varying covariates corresponding to the segmented data. For example, feature extraction can be performed on the public opinion data contained in the segmented data in the segmented dataset to obtain public opinion features; feature extraction can be performed on the regulatory data contained in the segmented data in the segmented dataset to obtain risk features; feature extraction can be performed on the logistics data contained in the segmented data in the segmented dataset to obtain logistics features; next, feature normalization processing is performed on the public opinion features, risk features, and logistics features to obtain the time-varying covariates corresponding to the segmented data in the segmented dataset.

[0034] Step 103: Use a random survival forest to predict the current time-varying covariates and obtain the current predicted cumulative risk.

[0035] The current forecast of cumulative risk is associated with medical devices and / or their suppliers.

[0036] In some embodiments, a random survival forest may include K survival trees, and the K survival trees may be pre-constructed based on a set of sample data related to medical devices; wherein K is an integer greater than 1 and may be used to characterize the number of survival trees contained in the random survival forest.

[0037] In some embodiments, the currently predicted cumulative risk may include the degree of risk that the procurement process of medical devices may face in future periods within the current time period, as characterized by current dynamic data; correspondingly, the currently predicted cumulative risk can be obtained in the following ways: The m-th time-varying covariate in the current time-varying covariate is classified and processed using the k-th survival tree in the random survival forest to determine the k-th target leaf node in the k-th survival tree corresponding to the m-th time-varying covariate. Then, the m-th time-varying covariate is processed using the k-th risk function corresponding to the k-th target leaf node in the k-th survival tree to obtain the k-th predicted risk. Next, the first predicted risk to the k-th predicted risk are statistically averaged to obtain the m-th predicted risk corresponding to the m-th time-varying covariate. Finally, the first predicted risk corresponding to the first time-varying covariate in the current time-varying covariate and the M-th predicted risk corresponding to the M-th time-varying covariate in the current time-varying covariate are statistically averaged to obtain the current predicted cumulative risk.

[0038] For example, the first predicted risk may include the result of classifying the m-th time-varying covariate through the first survival tree in the random survival forest, and the K-th predicted risk may include the result of classifying the m-th time-varying covariate through the K-th survival tree in the random survival forest.

[0039] For example, the risk function associated with the leaf nodes of the survival tree in a random survival forest can also be pre-constructed based on a sample dataset.

[0040] Step 104: Using the conditional survival probability prediction function of the Cox model, dynamically process the current predicted cumulative risk and the current time-varying covariate to perceive the current predicted risk probability associated with the medical device and / or supplier.

[0041] In some embodiments, the current predicted risk probability may include the procurement risk of medical devices predicted in the current time period that corresponds to a future time and is associated with the current predicted cumulative risk and the current time-varying covariate.

[0042] In some embodiments, the current predicted risk probability can be perceived in the following ways: The Cox model calculates the current predicted cumulative risk and the current time-varying covariate to obtain the current predicted survival probability, and determines the current predicted risk probability as the difference between 1 and the current predicted survival probability; specifically as shown in equation (1): (1) in, To predict the current survival probability, For the current predicted risk probability; This is the conditional survival probability prediction function for the Cox model. For the current time-varying covariate, The current cumulative risk can be predicted using the risk prediction function of the random forest model. The time-varying covariates are calculated as follows: It can represent the current moment, or it can represent the end time of the observation period. Representing future moments of the current time period, This is the baseline risk function, describing the instantaneous risk level when all covariates are zero. Represents any time value between the current time and a future time. and These are the model parameters of the Cox model, used to measure the log-linear contribution of current dynamic data and current predicted cumulative risk to the predicted survival probability, respectively.

[0043] Step 105: Determine the procurement decision for medical devices based on the current predicted risk probability.

[0044] The procurement decision includes risk tracing information for at least a portion of the current dynamic data.

[0045] In some embodiments, the procurement decision may include the type of medical device to be procured, the quantity to be procured, the procurement time, the procurement channel, and the device arrival time; wherein, the device arrival time is used to characterize the expected time for the medical device to reach the clinical and / or outpatient use status of the hospital.

[0046] Accordingly, procurement decisions can be determined in the following ways: If the current predicted risk probability is greater than or equal to the first risk threshold, the procurement decision can be determined as follows: reduce or not procure the above-mentioned medical devices, and select medical devices from the candidate devices whose predicted risk probability is less than or equal to the target threshold to replace the procurement of the above-mentioned medical devices; if the current predicted risk probability is less than the second risk threshold, the procurement decision can be determined as follows: procure medical devices according to the preset procurement schedule; wherein, the first risk threshold may be greater than the second risk threshold.

[0047] In some embodiments, risk tracing information may include at least a portion of the current dynamic data, and the degree of impact on procurement decisions or the current predicted risk probability; accordingly, risk tracing information may be obtained in the following ways: The adjustment step is determined, and the first data in the current time-varying covariate is adjusted based on the adjustment step to obtain the first covariate. The conditional survival probability prediction function of the Cox model is then used to calculate the first covariate and the current predicted cumulative risk to obtain the first calculation result. Then, the second data in the current time-varying covariate is adjusted based on the adjustment step to obtain the second covariate. The conditional survival probability prediction function of the Cox model is then used to calculate the second covariate and the current predicted cumulative risk to obtain the second calculation result. At this point, the first difference between the first calculation result and the current predicted survival probability is determined, and the second difference between the second calculation result and the current predicted survival probability is determined. If the first difference is greater than the second difference, it can be characterized that the first degree of influence represented by the risk tracing information corresponding to the first data is greater than the second degree of influence represented by the risk tracing information corresponding to the second data. By performing recursive traversal calculations on the data contained in the current time-varying covariate in the above manner, the risk tracing information corresponding to each data in the current time-varying covariate can be obtained.

[0048] As can be seen from the above, the intelligent decision-making and traceability method for medical device procurement based on dynamic perception provided in this application collects current dynamic data, which includes basic supplier data, public opinion data, regulatory data, and logistics data associated with the medical device in the current time period. Thus, through this current dynamic data, the actual status of the medical device in the supplier, public opinion, regulatory, and logistics dimensions can be comprehensively and diversely represented in the current time period. Furthermore, based on the current dynamic data, a current time-varying covariate is generated. This current time-varying covariate can more realistically reflect the pattern of data change over time in the current dynamic data. On this basis, a random survival forest is used to predict the current time-varying covariate to obtain the current predicted cumulative risk. This current predicted cumulative risk is associated with the medical device and / or its supplier. Thus, by using the random survival forest, the importance of features can be quantified and key risk factors can be identified. This approach offers advantages such as dynamic, accurate, and interpretable risk prediction, improving the accuracy of current cumulative risk prediction. Furthermore, by using the conditional survival probability prediction function of the Cox model, the current cumulative risk and current time-varying covariates are dynamically processed to perceive the probability of current predicted risk associated with medical devices and / or suppliers. Thus, leveraging the dynamic risk tracking advantage of the Cox model, the accuracy of current predicted risk probability can be improved. Moreover, procurement decisions for medical devices are determined based on the current predicted risk probability, and these decisions include risk tracing information from at least a portion of the current dynamic data. This not only achieves fine-grained risk tracing of at least a portion of the current dynamic data influencing procurement decisions but also enables the pre-determination of procurement decisions for medical devices. This overcomes the shortcomings of delayed adjustments in related technology procurement decisions and meets the actual procurement needs for medical devices.

[0049] Based on the foregoing embodiments, the intelligent decision-making and traceability method for medical device procurement based on dynamic perception provided in this application can be implemented by determining the procurement decision for medical devices based on the current predicted risk probability in the following ways: Step A1: Dynamically track and collect future dynamic data.

[0050] The future dynamic data includes supplier basic data, public opinion data, regulatory data, and logistics data for future time periods; the future time period includes the time period after the future moment of the current time period.

[0051] In some embodiments, the method for collecting current dynamic data as described above can be used to continuously track and collect supplier basic data, public opinion data, regulatory data, and logistics data in future time periods, thereby obtaining future dynamic data.

[0052] Step A2: Determine the probability of future risks corresponding to future dynamic data.

[0053] In some embodiments, the method provided in the foregoing embodiments can be used to generate future time-varying covariates corresponding to future dynamic data. Then, the future time-varying covariates can be predicted using a random survival forest to obtain the future predicted cumulative risk. Finally, the conditional survival probability prediction function of the Cox model can be used to dynamically process the future predicted cumulative risk and the future time-varying covariates to obtain the future predicted risk probability.

[0054] Step A3: Determine the procurement decision based on the current predicted risk probability and the future predicted risk probability.

[0055] In some embodiments, procurement decisions can be determined in the following ways: If both the current predicted risk probability and the future predicted risk probability are greater than or equal to the risk threshold, then the procurement decision can be determined as follows: reduce the procurement of medical devices, or not procure medical devices and select medical devices from the candidate devices whose predicted risk probability is less than or equal to the target threshold to replace the procurement of the aforementioned medical devices.

[0056] As can be seen from the above, the intelligent decision-making and traceability method for medical device procurement based on dynamic perception provided in this application dynamically tracks and collects future dynamic data. This future dynamic data includes supplier basic data, public opinion data, regulatory data, and logistics data for future time periods. Future time periods include time periods after the current time period. Thus, continuous and dynamic tracking and collection of supplier basic data, public opinion data, regulatory data, and logistics data can be achieved in the time dimension. Furthermore, the predicted risk probability corresponding to the future dynamic data is determined. This enables continuous prediction of the risk probabilities of supplier basic data, public opinion data, regulatory data, and logistics data corresponding to the current and future time periods in the time dimension. Based on this, procurement decisions are determined based on the current and future predicted risk probabilities, improving the targeting and accuracy of procurement decisions in the time dimension.

[0057] Based on the foregoing embodiments, the intelligent decision-making and traceability method for medical device procurement based on dynamic perception provided in this application can also perform the following operations: If the procurement decision includes early warning procurement information, the risk contribution score of the data in the current dynamic data to the current predicted risk probability is determined based on the first regression coefficient of the conditional survival probability prediction function and the current dynamic data; based on the risk contribution score, risk tracing information is generated.

[0058] The first regression coefficient includes the regression coefficients associated with time-varying covariates in the Cox model.

[0059] Accordingly, if the procurement decision does not include early warning procurement information, the risk contribution score can be determined without a clear definition.

[0060] In some embodiments, the first regression coefficient may include .

[0061] In some embodiments, the early warning procurement information may include a notification indicating that the risk level of the procurement of medical devices is greater than or equal to a risk threshold.

[0062] In some embodiments, risk warning procurement can be determined in any of the following ways: If the current predicted risk probability is greater than or equal to the risk threshold, the current predicted risk level corresponding to the current predicted risk probability can be quantified, and early warning procurement information including the current predicted risk level can be output.

[0063] If the current predicted risk probability is greater than or equal to the risk threshold and the future predicted risk probability is greater than the current predicted risk probability, then an early warning procurement information message indicating that the risk level of medical devices is continuously increasing can be output.

[0064] In some embodiments, the risk contribution score may include the magnitude of the influence or contribution of data in the current time-varying covariate corresponding to the current dynamic data on the current predicted risk probability; for example, the risk contribution score corresponding to the m-th data in the current dynamic data can be determined in the following way: The m-th component of the first regression coefficient is multiplied with the m-th covariate corresponding to the m-th data in the current time-varying covariate to obtain the m-th result. Then, the m-th result is determined as the risk contribution score corresponding to the m-th data in the current dynamic data.

[0065] For example, the risk contribution scores corresponding to each data point in the current dynamic data can be integrated to obtain a set of risk contribution scores, and the set of risk contribution scores can be determined as risk tracing information.

[0066] Specifically, the risk contribution score corresponding to the m-th data point in the current dynamic data. It can be calculated using equation (2): (2) in, This is the m-th component in the first regression coefficient. It is the m-th covariate in the current time-varying covariate that corresponds to the m-th data.

[0067] It should be noted that the first regression coefficient can be presented in matrix form, and the number of its components can be the same as the number of covariates in the current time-varying covariates.

[0068] As can be seen from the above, in the intelligent decision-making and traceability method for medical device procurement based on dynamic perception provided in this application embodiment, if the procurement decision includes early warning procurement information, the risk contribution score of the data in the current dynamic data to the current predicted risk probability is determined based on the first regression coefficient of the conditional survival probability prediction function and the data in the current time-varying covariate. In this way, the degree of influence of the data in the current dynamic data on the current predicted risk probability can be accurately quantified. On this basis, risk traceability information is generated based on the risk contribution score, which can improve the comprehensiveness and accuracy of the risk traceability information.

[0069] Based on the foregoing embodiments, the intelligent decision-making and traceability method for medical device procurement based on dynamic perception provided in this application can also perform the following operations: In response to the processing of risk tracing information, procurement decisions are adjusted.

[0070] In some embodiments, the processing of risk tracing information can be input by the user, or by the control device or the host computer device; this application embodiment does not limit this.

[0071] In some embodiments, the processing of risk tracing information can be implemented in the following ways: A filtering operation is performed on the risk contribution scores contained in the risk tracing information to obtain the target score. For example, the risk contribution scores corresponding to logistics data in the risk tracing information can be filtered by type dimension, and the risk contribution score corresponding to logistics data in the risk tracing information can be determined as the target score. Alternatively, the risk tracing information can be filtered by value dimension based on the magnitude of the risk contribution scores in the risk tracing information, and the risk contribution score with the largest value in the risk tracing information can be determined as the target score.

[0072] In some embodiments, controlling procurement decisions can be achieved through any of the following methods: Obtain the risk threshold corresponding to the target score, and then adjust the risk level of the procurement decision based on the score difference between the risk threshold and the target score to obtain the adjusted procurement decision. For example, if the score difference is less than or equal to the first data, the risk level of the procurement decision can be adjusted to the first level; if the score difference is greater than the second data, the risk level of the procurement decision can be adjusted to the second level. Here, the first data can be less than the second data, and the risk level represented by the first level can be less than the risk level represented by the second level.

[0073] Pre-set strategy adjustment templates for different risk types. Then, call the corresponding strategy adjustment template according to the risk type corresponding to the target score. With the goal of minimizing procurement losses, adjust procurement decisions based on procurement costs, emergency procurement premiums, supply disruption losses, inventory holding costs, and execution costs of decision-making actions. Among them, risk types can include qualification risks and capacity risks corresponding to supplier basic data, public opinion risks corresponding to public opinion data, quality risks corresponding to regulatory data, and logistics risks corresponding to logistics data.

[0074] For example, the above two methods can be combined to achieve targeted control over procurement decisions. For instance, the procurement decision can be targeted based on the risk type corresponding to the target score and the score difference through a strategy control template. For example, if the risk type corresponding to the target score is logistics risk and the score difference is greater than 40%, the procurement decision can be controlled through the strategy control template to suggest increasing the safety stock of medical devices in proportion to the score difference. If the risk type corresponding to the target score also includes public opinion risk, the procurement decision can also be controlled as follows: initiate the qualification review of candidate suppliers.

[0075] As can be seen from the above, in the intelligent decision-making and traceability method for medical device procurement based on dynamic perception provided in this application embodiment, the procurement decision is adjusted in response to the processing operation of risk traceability information. In this way, dynamic, flexible and closed-loop adjustment of procurement decision is realized, thereby improving the flexibility of procurement decision control.

[0076] Based on the foregoing embodiments, in the intelligent decision-making and traceability method for medical device procurement based on dynamic perception provided in this application embodiment, before dynamically processing the current predicted cumulative risk and the current time-varying covariate through the conditional survival probability prediction function of the Cox model to perceive the current predicted risk probability associated with the medical device and / or supplier in the current time period, the following steps may also be performed: Step B1: Using the partial likelihood estimation method, based on the sample data set and the cumulative predicted risk of the sample, determine the first and second regression coefficients of the Cox model.

[0077] Among them, the first regression coefficient is associated with the time-varying covariates of the sample dataset; the second regression coefficient is associated with the time-varying derived features of the sample dataset; the cumulative prediction risk of the sample is obtained by processing the sample data in the sample dataset using a random survival forest; and the cumulative prediction risk of the sample is associated with the time-varying derived features of the sample.

[0078] In some embodiments, the sample dataset may include a first set and a second set, wherein the data in the first set and the data in the second set may have a one-to-one correspondence; for example, the first set may include supplier basic data, public opinion data, regulatory data and logistics data related to various medical device samples from multiple historical periods, while the second set may include sample time-varying covariates corresponding to the data in the first set respectively.

[0079] Specifically, dynamic historical data from multiple sources of medical device samples across multiple historical time periods can be collected. This dynamic historical data can then be integrated based on the medical device sample identifier or the supplier identifier of the medical device sample to identify risk events within the dynamic historical timeframe. For example, a survival time T can be defined for each supplier or medical device sample, starting from the beginning of the observation period. If a risk event occurs during the observation period, T represents the occurrence time of the risk event; otherwise, T represents the end time of the observation. For example, risk event indicator variables can also be defined. If the variable takes a value of 1, it indicates that a risk event has occurred; if the variable takes a value of 0, it indicates that no risk event has occurred. At this time, a first set can be obtained, which includes event description data, survival time, risk event indicator variables, and supplier identifiers corresponding to the risk event. Among them, the event description data can include supplier basic data, public opinion data, regulatory data, and logistics data.

[0080] For example, risk events may include major quality incidents, product recalls, supply disruptions exceeding a threshold, or supplier qualification failures.

[0081] For example, the method for generating current time-varying covariates based on current dynamic data provided in the foregoing embodiments can be used to process the corresponding data in the first set to obtain sample time-varying covariates in the second set.

[0082] Specifically, the data corresponding to the same supplier in the first set can be segmented according to a fixed time window. For example, the j-th segment time corresponding to the i-th supplier in the first set can be segmented by... This indicates that, within the aforementioned time segments, the observation time can be... The start time of the observation can be The end time of the observation period can be .

[0083] For example, the dynamic characteristics of the data within the observation period can be used to obtain sample covariates corresponding to each time segment. Then, according to the order of the time segments, the sample covariates are integrated to obtain the sample covariates corresponding to a single supplier. The sample covariates corresponding to each supplier are then integrated to obtain the sample time-varying covariates corresponding to all suppliers. At this point, a second set can be obtained.

[0084] For example, the data ultimately used to determine the first regression coefficient and the second regression coefficient can be a second set; for example, the second set can be denoted as... ;in, Identify the nth supplier. For the nth sample, the time-varying covariate, The start time of the time interval. The moment a risk event occurs. Let N be the indicator variable for the nth risk event corresponding to the time-varying covariate of the nth sample. N is greater than 1 and is used to represent the total number of samples in all time intervals. n is an integer greater than or equal to 1 and less than or equal to N.

[0085] In some embodiments, the first regression coefficient can be The second regression coefficient can be .

[0086] In some embodiments, the cumulative risk of sample prediction can be obtained in the following way: The sample data in the sample dataset is input into the k-th survival tree in the random survival forest. The n-th sample data is classified by the k-th survival tree to obtain the k-th leaf node corresponding to the n-th sample data in the k-th survival tree. The cumulative risk function associated with the k-th leaf node is used to calculate the predicted risk of the k-th sample corresponding to the n-th sample data. Next, all survival trees in the random survival forest are traversed using the above method to obtain the predicted risks of K samples. The cumulative predicted risk corresponding to the n-th sample is obtained by statistically averaging the predicted risks of the K samples. Finally, the cumulative predicted risks corresponding to all samples are integrated to obtain the cumulative predicted risk of the sample. Specifically, it can be shown in equation (3): (3) in, To predict the risk of the k-th sample corresponding to the n-th sample data, The cumulative predicted risk for the nth sample data is specifically used to characterize the time-varying covariance of the nth sample and the start time of that interval. Under these conditions, random survival forest is used from the observation starting point. At the designated time The calculated cumulative risk value reflects the impact of the random survival forest on the k-th sample data over time. The comprehensive estimate of cumulative risk, where K is an integer greater than 1.

[0087] In some embodiments, the time-varying derived features of the samples can vary depending on the different sample data in the sample dataset and the different time periods corresponding to the sub-data in the sample data. These features can characterize the level of procurement risk represented by the sample data in the sample dataset. For example, the time-varying derived features of the nth sample data can include the cumulative predicted risk corresponding to the nth sample data at time [time value missing]. The value of is used to characterize up to The cumulative risk level is assessed based on the sample characteristics of the nth sample data; specifically, it can be characterized by equation (4): (4) in, The time-varying derived features of the nth sample data.

[0088] In some embodiments, the first regression coefficient and the second regression coefficient can be determined in the following manner: First, a proportional hazards model of the Cox model is constructed, and the hazard function of this model over the time interval is given. This can be expressed by equation (5): (5) in, This is the baseline risk function for the Cox model, used to calculate the baseline risk when all time-varying covariates are 0. and The first and second regression coefficients are to be estimated, used to measure the log-linear contribution of the time-varying covariates and time-varying derived features of the sample to the calculation results of the risk function, respectively. Used for Perform transpose calculation.

[0089] Secondly, a partial likelihood function suitable for the counting process is constructed, as shown in equation (6): (6) The above formula is used to maximize the estimated parameters. and The partial likelihood function compares the risk scores of the k-th sample data when the event occurs. With all sample data in the sample dataset Sum of risk scores To maximize the conditional probability of the event, the Newton-Raphson algorithm is used to solve for the estimated value of the first regression coefficient. Estimates of the second regression coefficient .

[0090] in, It is the first The end time of the observation time interval corresponding to the sample data; For the k-th sample data in The risk set corresponding to a given moment is included in... The set of data in the k-th sample data that is still under observation but has not yet occurred, i.e., satisfying samples .

[0091] Next, the baseline cumulative risk function can be estimated using the expanded form of the counting process estimated by Breslow, as shown in Equation (7): (7) in, The estimated baseline cumulative risk function is a step function that jumps in value at the moment a risk event occurs. The magnitude of this jump is determined by the cumulative predicted risk at that moment. For The indicator variable for risk events within the time period ending at the specified time.

[0092] Step B2: Construct a conditional survival probability prediction function based on the baseline risk function, first regression coefficient, and second regression coefficient of the Cox model.

[0093] For example, the conditional survival probability prediction function can be shown as Equation (1).

[0094] As can be seen from the above, in the intelligent decision-making and traceability method for medical device procurement based on dynamic perception provided in this application embodiment, the first regression coefficient and the second regression coefficient of the Cox model are determined by the partial likelihood estimation method based on the sample data set and the sample cumulative prediction risk. The first regression coefficient is associated with the sample time-varying covariates of the sample data set, and the second regression coefficient is associated with the sample time-varying derived features of the sample data set. The sample cumulative prediction risk is obtained by processing the sample data in the sample data set by random survival forest, and the sample cumulative prediction risk is associated with the sample time-varying derived features. In this way, the accuracy of the first regression coefficient and the second regression coefficient can be improved. Furthermore, based on the baseline risk function of the Cox model, the first regression coefficient, and the second regression coefficient, a conditional survival probability prediction function is constructed, which can improve the relevance of the conditional survival probability prediction function.

[0095] Based on the foregoing embodiments, the intelligent decision-making and traceability method for medical device procurement based on dynamic perception provided in this application can also perform the following operations: A random survival forest is constructed based on the sample dataset; the cumulative predicted risk of the samples is calculated by using the risk function associated with the leaf nodes of the survival trees in the random survival forest.

[0096] In some embodiments, a random survival forest can be constructed in the following ways: First, using the Random Survival Forest algorithm, sampling with replacement is performed on the data contained in the second set of the sample dataset, using supplier identifiers as units, to generate B bootstrap sample sets. In other words, all time interval records of a single supplier are sampled as a whole, ensuring that longitudinal observations of the same supplier may or may not appear simultaneously in the bootstrap samples, thereby reducing the risk of data leakage. Secondly, for each bootstrap sample set, a survival tree is constructed; specifically, each node can be randomly selected. For each candidate feature, calculate the log-rank test statistic at different split points. Select the feature and split point with the largest statistic for splitting, and recursively split until the number of risk events within a node is too small or the minimum number of samples in the node is reached; where, usually p can be an integer greater than 1 and is used to represent the total number of features possessed by each node. The above statistics can be determined based on the start and end times of each time interval and the risk event indicator variable.

[0097] Next, after each individual survival tree is generated, for any data falling into the second set of leaf nodes of any survival tree, the risk function of that leaf node can be calculated using the Nelson-Aalen estimation suitable for the counting process, i.e., by applying the risk function to all data falling into the second set of leaf nodes at time... All events occurring within the previously concluded interval are weighted and summed. The risk function of each tree is used to obtain the cumulative risk prediction for each sample (each time interval), as shown in Equation (3).

[0098] As can be seen from the above, the intelligent decision-making and traceability method for medical device procurement based on dynamic perception provided in this application constructs a random survival forest based on a sample dataset. The cumulative predicted risk of the samples is calculated using the risk function associated with the leaf nodes of the survival trees in the random survival forest. This improves the relevance of the random survival forest, thereby increasing the accuracy of the risk function and the accuracy of the cumulative predicted risk of the samples.

[0099] In order to quantify the technical effects of the technical solution provided in this application, this application also provides a data comparison of the execution results of different solutions.

[0100] In the above data comparison process, a test environment was built based on real historical procurement data and simulated high-risk scenarios to verify the performance advantages of this solution in dynamic risk perception and forward-looking early warning. Specifically, the historical procurement data may include the medical device procurement database of a tertiary hospital from 2019 to 2023, which contains complete records of 120 suppliers and 50 types of high-risk consumables, and synthetic risk events are injected through the Monte Carlo method to expand the extreme scenario sample.

[0101] For example, risk events are specifically defined; specifically, risk events may include sudden public opinion events, frequent regulatory notifications, and abnormal logistics disruptions; among them, sudden public opinion events may include a jump in the negative public opinion score of a supplier from 0.2 to 0.8 within 1 day, simulating a major quality exposure; frequent regulatory notifications may include an increase in the number of notifications to a supplier from 0 to 5 within 3 days; and abnormal logistics disruptions may include a sudden increase in the logistics delay rate from 2% to 15%.

[0102] For example, during the collection of historical procurement data, a dynamic adjustment mechanism can be used to adjust the forecast period; specifically, the risk probability can be updated every 24 hours.

[0103] The comparison schemes can include Scheme 1 through Scheme 4. Scheme 1 relies on static annual audits of supplier qualifications to determine procurement decisions; Scheme 2 is based on real-time monitoring of data with fixed thresholds and adjusts procurement decisions according to the monitoring results; Scheme 3 uses a Cox proportional hazards model to analyze static characteristics to determine procurement decisions; and Scheme 4 uses a random survival forest to analyze static characteristics and determine procurement decisions based on cumulative risk. Table 1 presents the statistical results of the actual application effects of the comparative schemes provided in the embodiments of this application and the scheme itself.

[0104] As shown in Table 1, this scheme is significantly better than the comparison scheme in terms of warning accuracy, recall, warning lead time, false alarm rate, and single inference time. Compared with the third scheme, this scheme improves the warning accuracy by 14.5%, the recall by 23.3%, the warning lead time by 3.3 days, the false alarm rate by 39%, and the single inference time is also shortened.

[0105] In the process of comparing the data with the above-mentioned comparative schemes, the robustness of this scheme was also verified. Under the extreme risk intensity of 0.9, that is, in the case of a major public opinion crisis, the early warning time of this scheme remains at 7.8 days, while the early warning time of the fourth scheme is reduced to 3.5 days and the early warning time of the second scheme is reduced to 1.2 days.

[0106]

[0107] Table 1 On the other hand, in order to verify the independent contribution of each data processing step in this scheme to the data processing effect, this application also provides the following ablation experiments: Removing the data processing step for constructing the current time-varying covariate in this solution reduces the recall rate to 78.2%. Removing the data processing step for the current predicted cumulative risk probability associated with the random forest model in this solution changes the early warning lead time to 6.1 days. Removing the data processing step for processing the current predicted cumulative risk probability and the current time-varying covariate using the Cox model in this solution causes the early warning accuracy to drop to 78.5% (-8%) and the false alarm rate to rise to 6.7% (+60%). However, modifying the conditional survival probability prediction function of the Cox model from integral form to discrete numerical calculation reduces the early warning lead time to 8.1 days (-15%) and the recall rate to 84.2% (-5%). Removing the data processing step for dynamically tracking and collecting future dynamic data and determining the future predicted risk probability, and then determining procurement decisions based on the current predicted risk probability and the future predicted risk probability, reduces the early warning lead time to 7.2 days (-24%) and the recall rate to 83.5% (-6%).

[0108] In summary, the collaborative contribution of the data processing stage related to the current predicted cumulative risk probability of the random survival forest and the current time-varying covariates, as well as the data processing stage related to the Cox model, to the data processing effect of this scheme are significant.

[0109] The description of the various embodiments above tends to emphasize the differences between the various embodiments. The similarities or similarities between them can be referred to, and for the sake of brevity, they will not be repeated here.

[0110] The methods disclosed in the various method embodiments provided in this application can be arbitrarily combined to obtain new method embodiments without conflict.

[0111] The features disclosed in the various product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.

[0112] The features disclosed in the various method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method or device embodiments.

[0113] It should be noted that the aforementioned computer-readable storage media can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM), etc.; or it can be various electronic devices including one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0114] It should be noted that, in this document, 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 a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0115] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0116] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware nodes. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0117] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0118] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0119] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0120] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for intelligent decision-making and traceability in medical device procurement based on dynamic perception, characterized in that, include: Collect current dynamic data; wherein, the current dynamic data includes at least the basic data of the suppliers associated with the medical device, public opinion data, regulatory data, and logistics data for the current time period; Based on the current dynamic data, generate the current time-varying covariate; The current time-varying covariate is predicted using a random survival forest to obtain the current predicted cumulative risk; wherein the current predicted cumulative risk is associated with the medical device and / or the supplier of the medical device. The current predicted cumulative risk and the current time-varying covariate are dynamically processed using the conditional survival probability prediction function of the Cox model to perceive the current predicted risk probability associated with the medical device and / or the supplier. The procurement decision for the medical device is determined based on the current predicted risk probability; wherein the procurement decision includes risk tracing information for at least a portion of the current dynamic data.

2. The method according to claim 1, characterized in that, The process of determining the procurement decision for the medical device based on the current predicted risk probability includes: Dynamically track and collect future dynamic data; wherein, the future dynamic data includes supplier basic data, public opinion data, regulatory data, and logistics data for future time periods; the future time period includes the time period after a future moment of the current time period; Determine the probability of future predicted risks corresponding to the aforementioned future dynamic data; The procurement decision is determined based on the current predicted risk probability and the future predicted risk probability.

3. The method according to claim 1 or 2, characterized in that, The method further includes: If the procurement decision includes early warning procurement information, the risk contribution score of the data in the current dynamic data to the current predicted risk probability is determined based on the first regression coefficient of the conditional survival probability prediction function and the data in the current time-varying covariate; wherein, the first regression coefficient includes the regression coefficient in the Cox model associated with the time-varying covariate; The risk tracing information is generated based on the risk contribution score.

4. The method according to claim 1 or 2, characterized in that, The method further includes: The procurement decision is adjusted in response to the processing of the risk tracing information.

5. The method according to claim 1, characterized in that, Before dynamically processing the current predicted cumulative risk and the current time-varying covariate using the conditional survival probability prediction function of the Cox model to perceive the current predicted risk probability associated with the medical device and / or the supplier, the method further includes: Using partial likelihood estimation, based on the sample dataset and the cumulative prediction risk, the first regression coefficient and the second regression coefficient of the Cox model are determined. The first regression coefficient is associated with the time-varying covariates in the sample dataset; the second regression coefficient is associated with the time-varying derived features of the sample dataset; the cumulative prediction risk is obtained by processing the sample data in the sample dataset using the random survival forest; and the cumulative prediction risk is associated with the time-varying derived features. Based on the baseline risk function of the Cox model, the first regression coefficient, and the second regression coefficient, the conditional survival probability prediction function is constructed.

6. The method according to claim 5, characterized in that, The method further includes: The random survival forest is constructed based on the sample data set; The cumulative predicted risk of the sample is calculated using the risk function associated with the leaf nodes of the survival tree in the random survival forest.

Images