Purchase decision optimization method for easy-to-expire medicines in hospitals

Abstract

The invention relates to the technical field of medicine purchasing, and provides a hospital easy-to-expire medicine purchasing decision optimization method, which comprises the following steps of: firstly, extracting medicine inventory, consumption records and purchasing management parameters in a hospital information system, and constructing a complete inventory state data set; then, calling a demand prediction model constructed based on a time sequence and external factors, and outputting drug prediction demand values in a plurality of periods in the future; thirdly, constructing a multi-cycle purchase optimization model according to the current inventory and the prediction demand; then, dividing a planning period into rolling windows, and gradually generating an ordering scheme meeting conditions by using a heuristic greedy and local search algorithm; then, the scheme is pushed to a purchase decision-making interface, and a formal order is submitted and generated after the scheme is confirmed by a manager; after a supplier delivers goods and stores the goods, the system updates inventory information and records scrapping and shortage data; and finally, adjusting the model through a feedback mechanism so as to improve the accuracy of subsequent prediction and decision.

Description

Technical Field

[0001] The present invention relates to the technical field of drug procurement, and more specifically, to a hospital's method for optimizing procurement decisions for easily expired drugs. Background Art

[0002] Hospital drug inventory management is crucial for ensuring healthcare quality and operational efficiency, especially for expired drugs (drugs with a short shelf life that are easily lost due to unused inventory). Inappropriate purchasing decisions often lead to inventory overstocks and expired drugs, resulting in wasted resources. According to statistics, studies indicate that approximately 20% of hospital drug inventory is ultimately discarded due to expiration, resulting in losses of up to $33 billion. Expired drugs are particularly serious within hospital supply chains, negatively impacting both hospital finances and patient care.

[0003] For example, in a system dynamics modeling study of 16 hospital pharmacies in Iran, literature 1 (Shahbahrami E, et al., A dynamic management model for sustainable drug supply chain in hospital pharmacies in Iran. BMC Health ServRes, 2024) found that if drug inventory management is not combined with dynamic information flow and human factor optimization, problems such as drug scrapping, retention of low-consumption drugs, and low inventory turnover will continue to occur, ultimately affecting pharmacy profits and inventory efficiency.

[0004] At the same time, inventory shortages are also a prominent issue: certain specialty drugs often experience shortages during periods of temporary demand peaks. The coexistence of overstocking and shortages reflects the inability of existing procurement management methods to accurately balance supply and demand. Many hospitals still rely on traditional manual experience for procurement and inventory management, lacking scientific forecasting and optimization methods, resulting in information lags and untimely decision-making. Reference 2 (Mettler T, Rohner P., E-Procurement in hospital pharmacy: An exploratory multi-case study from Switzerland. J Theor Appl Electron Commer Res, 2009) notes that in some small and medium-sized hospitals in Switzerland, the drug procurement process relies on paper orders and manual data transfer. Pharmacies lack visibility into supplier status and lack effective electronic support systems, severely hindering the modernization of procurement management.

[0005] Because of this, drug expiration and waste occur repeatedly in many hospitals, seriously affecting inventory turnover and capital utilization efficiency.

[0006] To address the uncertainty in drug procurement decisions, some research has begun to explore the use of machine learning and optimization methods. For example, Reference 3 (Chung TH, et al., Improving access to essential medicines in Sierra Leone via machine learning, 2024) proposes combining multi-task learning with stochastic optimization to forecast drug demand in the medical supply chain and minimize the cost of "unmet demand" based on the forecast results. However, while such methods have shown some success in practical deployment, they primarily focus on the fairness and accessibility of overall drug distribution and have yet to provide customized modeling and constrained optimization for the procurement of easily expired drugs.

[0007] Therefore, although existing technologies have initially introduced intelligent methods to improve the efficiency of medical resource allocation, there are still obvious deficiencies in the detailed modeling, constraint processing and system integration for the procurement of easily expired drugs. There is an urgent need for a method that can be embedded in the actual procurement process and takes into account both prediction accuracy and decision optimization effects. Summary of the Invention

[0008] To overcome the aforementioned shortcomings of the prior art, the present invention provides a method for optimizing hospital procurement decisions for easily expired drugs. This method sequentially comprises data extraction, demand forecasting based on weighted losses, multi-period procurement optimization modeling with shelf life and storage capacity constraints, and a solution process using heuristic greedy and local search algorithms within a rolling window. The method uses drug expiration costs and shortage penalties as the core optimization objectives, combining inventory evolution with supply cycle constraints to dynamically generate per-period ordering recommendations. This method effectively addresses the existing issues of reliance on experience, predictions divorced from decision-making, and the inability to precisely control obsolescence risk.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] A hospital's method for optimizing procurement decisions for easily expired drugs includes the following steps:

[0011] Step S1: Extract the latest inventory data, drug consumption data, and procurement management-related parameters from the hospital information system. The inventory data includes the current inventory quantity of each drug, the time each batch entered the warehouse, and early warning information for drugs approaching expiration. The drug consumption data includes recent records of departmental withdrawals. The procurement management-related parameters include the purchase price, shelf life, supplier delivery cycle and batch requirements, remaining warehouse capacity, and budget status of each drug.

[0012] Step S2: Calling a drug demand forecasting model to forecast the demand for each drug for several future cycles. The drug demand forecasting model reweights the training data to match the loss function of the machine learning model with the decision optimization goal, assigning higher penalty weights to overestimated predictions that result in drug expiration and waste, and underestimated predictions that result in drug shortages.

[0013] Step S3: Generate the initial state of the current optimization cycle, including the estimated ending inventory of each drug if no purchase is made during this cycle and the quantity that may expire and be scrapped;

[0014] Step S4: construct a drug procurement optimization model and set constraints for the drug procurement optimization model. The objective function of the optimization model is:

[0015]

[0016] Where, is the purchase cost coefficient of drug i, is the unit inventory holding cost coefficient, is the unit cost of expired goods (which can be equal to the drug cost or higher to reflect the implicit loss), is the unit shortage penalty cost (such as the replacement cost or potential treatment delay loss caused by drug shortage), x i,t is the quantity of drug i ordered in period t, I i,t is the effective quantity of drug i in inventory at the end of period t, E i,t is the number of expired and scrapped drugs i in period t, S i,t is the shortage quantity, T is the total number of planning cycles, i∈I is any drug i in the drug set I;

[0017] Step S5: Divide the entire planning cycle into shorter rolling windows, solve the simplified subproblem in each window and obtain the decision of the current cycle;

[0018] Step S6, executing a heuristic greedy and local search algorithm, including generating an initial feasible solution based on a reorder point strategy; locally adjusting the drug order quantity in the initial solution, reducing the order quantity of drugs whose expected inventory exceeds a threshold, and increasing the order quantity of drugs whose expected shortage is greater than zero; and performing multiple rounds of iterative optimization until the improvement in the objective function value is less than a preset threshold or the maximum number of iterations is reached, while satisfying the warehouse capacity constraint;

[0019] Step S7: The solution generated by the optimization algorithm is presented on the purchasing decision support interface, listing the current inventory, predicted consumption, and recommended order quantity for each drug;

[0020] Step S8: Submit the confirmed order plan, generate a formal purchase order, and place the order with the supplier;

[0021] Step S9: After the supplier delivers the medicine and the medicine is put into storage, the inventory database is updated to register the date of entry and expiration date of the medicine;

[0022] Step S10, record the types and quantities of expired and scrapped drugs in the previous cycle, and adjust the drug demand forecast model and optimization parameters through the feedback mechanism.

[0023] As a further solution of the present invention, the objective function comprehensively considers four cost factors: the first is the procurement cost of the drug, which is used to measure the economic expenditure of the procurement itself; the second is the inventory holding cost, which reflects the storage resources and capital costs occupied by the drug while in stock; the third is the expiration and scrapping cost, which is used to punish the losses caused by the drug not being used within the validity period; and the fourth is the shortage penalty cost, which is used to evaluate the impact of the drug's failure to supply on time on clinical use. By weighting and summarizing these various costs, an objective function for procurement optimization is constructed, which serves as the optimization basis for the algorithm solution, thereby achieving procurement decisions that minimize overall costs.

[0024] As a further solution of the present invention, the constraints of the drug procurement optimization model include:

[0025] Demand satisfaction constraints: For each cycle and drug, the sum of effective inventory and current purchases must meet demand, and the allowed unmet demand is counted as a shortage S i,t , and is penalized in the target, i.e. I i,t +x i,t +S i,t ≥D i,t , and S i,t ≥0, where D i,t To forecast demand;

[0026] Inventory evolution constraint: the inventory at the end of the current cycle is equal to the inventory of the previous cycle plus the current purchase minus the demand consumption, and negative inventory is not allowed. At the same time, it is subject to the storage capacity limit, that is, in is the storage capacity limit of drug i, (·) + Indicates that the lower limit is 0 (negative inventory is not allowed, that is, unmet demand can be counted as a shortage S i,t =maxD i,t -I i,t -x i,t ,0);

[0027] Inventory capacity constraint: the total inventory of all drugs in each cycle must not exceed the total capacity of the warehouse to prevent blind purchases that exceed the storage capacity, that is, in is the total warehouse capacity;

[0028] Procurement cycle and supply constraints: Drug orders can only be placed within the permitted procurement cycle, and the impact of supplier delivery lead time must be considered, i.e., x i,t It can only be taken within the allowed procurement cycle. If the supply lead time is cycle, then decide when making a decision Inventory replenishment after the cycle.

[0029] Other business constraints: such as procurement budget limits (purchase expenditure per cycle Not exceeding the budget), supplier batch restrictions (some drugs must be purchased in boxes at a time, such as x i,t is a specific multiple), etc., can be added to the model as needed.

[0030] As a further solution of the present invention, the heuristic greedy + local search algorithm includes:

[0031] Initially, a feasible solution is generated based on empirical rules. Based on the (s, S) strategy, a reorder point and replenishment limit are set for each drug. If the inventory is lower than the reorder point, the drug is ordered to the replenishment limit.

[0032] Make local adjustments to the solution to reduce costs, identify drugs whose expected inventory after ordering exceeds a preset threshold, and try to reduce their order quantity to reduce expiration costs;

[0033] Increase order quantities for drugs with projected shortages greater than zero until the cost of the shortage penalty balances the cost of increased inventory;

[0034] When it is found that the total inventory exceeds the warehouse capacity limit, the order quantity of some drugs is reduced in ascending order according to the expiration date of the drugs so that the total inventory meets the capacity constraint.

[0035] Through multiple rounds of iteration, until the improvement of the objective function value is less than the preset threshold or the maximum number of iterations is reached.

[0036] As a further solution of the present invention, the drug demand forecasting model is constructed based on an LSTM network of time series; the input features of the model include time features, trend features and external event features in the historical drug consumption data; the time features include month, season and holiday identifiers; the trend features include moving averages of the past 3 to 12 cycles; the external event features include epidemic indicators or seasonal disease epidemic indexes; the model training process adopts a weighted loss function, and the loss function of the machine learning model is matched with the decision optimization goal by reweighting the training data according to the degree of impact on the decision cost. A first penalty weight is assigned to the overestimation of the prediction that causes drug expiration and waste, and a second penalty weight is assigned to the underestimation of the prediction that causes drug shortages. The first penalty weight and the second penalty weight are greater than the weight of the standard sample, so that the model training goal is consistent with the optimization goal of the procurement decision; the drug demand forecasting model outputs the demand forecast value and its 95% confidence interval for the next 4 to 12 procurement cycles.

[0037] As a further solution of the present invention, a hospital easily expired drug procurement management system includes:

[0038] The data collection and status monitoring module is used to extract the latest inventory data and drug consumption data from the hospital information system. The inventory data includes the current inventory quantity of each drug, the time each batch entered the warehouse, and the warning information of the approaching expiration of the drugs in stock. The drug consumption data includes the recent records of the amount of drugs issued by each department. The module is also used to obtain parameters related to procurement management, including the purchase price of each drug, shelf life, supplier delivery cycle and batch requirements, remaining warehouse capacity and budget status;

[0039] A demand forecasting module is used to call a drug demand forecasting model to predict the demand for each drug in the next Δ period. The drug demand forecasting model reweights the training data to match the loss function of the machine learning model with the decision optimization goal;

[0040] The procurement optimization decision calculation module is used to generate the initial state of the current optimization cycle, build a drug procurement optimization model, divide the entire planning cycle into shorter rolling windows, and execute a heuristic greedy + local search algorithm to obtain the recommended ordering plan for the current cycle;

[0041] The decision-making, approval, and execution module is used to present the solution provided by the optimization algorithm, generate a formal purchase order, and connect with the supplier to place the order;

[0042] The inventory update and feedback learning module is used to update the inventory database after the supplier delivers the goods and verifies them into the warehouse, record the types and quantities of expired and scrapped medicines, and feed this data back to the demand forecasting module and procurement optimization decision calculation module to adjust the optimization algorithm parameters.

[0043] As a further solution of the present invention, the drug inventory in the inventory data comes from the inventory management module of the hospital information system, the entry time of each batch is recorded in the entry document of the ERP system, and the expiration warning information is automatically generated according to the difference between the current date and the expiration date; the drug consumption data comes from the hospital department's collection record and the drug usage data of the HIS system; among the procurement management related parameters, the purchase price is extracted based on the procurement contract signed between the hospital and the supplier, the shelf life information comes from the drug instructions and packaging labels, and the supplier's delivery cycle and batch requirements come from the supplier agreement; the cost coefficient in the objective function is set as follows: the procurement cost coefficient is taken from the unit purchase price of the drug, the inventory holding cost coefficient is calculated at an annualized rate of 15%-25% of the drug value, the unit expired scrap cost is 1-2 times the drug cost to reflect the additional processing cost and opportunity loss, the shortage penalty cost is determined according to the clinical importance of the drug, and the procurement cost is 5-10 times for Class A (critical drugs), 3-5 times for Class B (important drugs), and 1-3 times for Class C (general drugs); in the drug demand forecasting model The time characteristics include the annual, monthly, and weekly data of the drug consumption time series and holiday identifiers, which are obtained from the calendar database of the hospital information system; the trend characteristics include the moving average of the past 3 to 12 cycles, which is calculated by sliding windowing the historical consumption data; the external event characteristics include the seasonal disease epidemic index obtained from the Centers for Disease Control and Prevention and the epidemic indicators obtained from the health department; the promotion adjustment coefficient in the processing of expiring drugs ranges from 0.1 to 0.3, which is determined by analyzing the effect of promotional activities on the increase in drug consumption in historical data; the safety stock is set at 1.5 to 2.5 times the average periodic demand for drugs plus the predicted standard deviation, and this ratio varies according to the level of importance of the drugs; the procurement budget limit is imported from the financial system or budget table.

[0044] Compared with the prior art, the method for optimizing hospital procurement decisions for easily expired drugs of the present invention has the following beneficial effects:

[0045] This invention directly links drug demand forecasts to the cost constraints in the optimization objective function, significantly different from the existing "forecast first, optimize later" process. By assigning different forecast error penalty weights to expiration waste and shortage risks, the forecast results are more closely aligned with actual decision-making objectives, thereby improving the effectiveness and fault tolerance of procurement plans. Compared with traditional models based on minimum mean square error training, this invention can effectively reduce the negative impact of forecast bias on inventory decisions and enhance the robustness of high-risk drug procurement.

[0046] This invention employs a solution strategy that combines heuristic greedy optimization with local search to generate a procurement plan within each procurement cycle that balances inventory turnover, obsolescence risk, and budget constraints. This approach differs from existing manual ordering rules based on static thresholds and can dynamically adjust ordering strategies to address cyclical fluctuations and sudden demand. Comparative testing has shown that the algorithm can significantly reduce obsolescence rates and warehouse backlogs, with particularly significant advantages for high-value pharmaceuticals with volatile demand, demonstrating enhanced practicality and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is the architecture diagram of the intelligent distribution system for the pharmaceutical supply chain based on the prediction-optimization collaborative mechanism in Reference 3.

[0048] Figure 2 This is the regional distribution map of medical intervention in Sierra Leone from Reference 3.

[0049] Figure 3 The figure is a flow chart of a method for optimizing hospital procurement decisions for easily expired drugs according to the present invention.

[0050] Figure 4 A schematic diagram of a rolling optimization window for a hospital's easily expired medicine procurement decision optimization method according to the present invention.

[0051] Figure 5 This is the user interface of a hospital's easily expired medicine procurement decision optimization method according to the present invention.

[0052] In the figure, Supply: Supply; Government Database (DHIS2): Government Database (DHIS2); Data Extraction & Processing: Data extraction and processing; Decision-Aware Prediction: Decision-aware prediction; Stochastic Optimization: Stochastic optimization; Picking from Warehouse: Picking from warehouse; Distribution to Facilities: Distribution to medical institutions; Consumption: Drug consumption. DETAILED DESCRIPTION

[0053] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0054] A hospital's method for optimizing procurement decisions for easily expired drugs includes the following steps:

[0055] Step S1, extracting the latest inventory data, drug consumption data and procurement management related parameters from the hospital information system. The inventory data includes the current inventory quantity of each drug, the time of each batch entering the warehouse, and the warning information of the expiring drugs in the warehouse. The drug consumption data includes the recent department’s collection quantity record; the procurement management related parameters include the purchase price of each drug, shelf life, supplier delivery cycle and batch requirements, warehouse remaining capacity and budget situation.

[0056] The drug inventory in the inventory data comes from the inventory management module of the hospital information system. The entry time of each batch is recorded in the entry document of the ERP system. The expiration warning information is automatically generated based on the difference between the current date and the expiration date. The drug consumption data comes from the hospital department's collection records and the drug usage data of the HIS system. Among the procurement management related parameters, the purchase price is extracted based on the procurement contract signed between the hospital and the supplier, the shelf life information comes from the drug instructions and packaging labels, and the supplier's delivery cycle and batch requirements come from the supplier agreement. The cost coefficient in the objective function is set as follows: the procurement cost coefficient is taken from the unit purchase price of the drug, the inventory holding cost coefficient is calculated at an annualized rate of 15%-25% of the drug value, the unit expired scrap cost is 1-2 times the drug cost to reflect the additional processing cost and opportunity loss, and the shortage penalty cost is determined according to the clinical importance of the drug. Grade A (critical drugs) takes the procurement cost of 5-1 0 times, Class B (important drugs) takes 3-5 times the purchase cost, and Class C (general drugs) takes 1-3 times the purchase cost; the time features in the drug demand forecasting model include the annual, monthly, weekly data of the drug consumption time series and holiday identifiers, which are obtained from the calendar database of the hospital information system; trend features include the moving average of the past 3 to 12 cycles, which is calculated by sliding window calculation of historical consumption data; external event features include the seasonal disease epidemic index obtained from the CDC and the epidemic indicators obtained from the health department; the promotion adjustment coefficient in the handling of expiring drugs ranges from 0.1 to 0.3, and is determined by analyzing the effect of promotional activities on the increase in drug consumption in historical data; the safety stock is set at 1.5 to 2.5 times the average periodic demand of the drug plus the predicted standard deviation, and this ratio varies according to the importance level of the drug; the procurement budget limit is imported from the financial system or budget table.

[0057] In step S2, a drug demand forecasting model is called to forecast the demand for each drug in several future cycles. The drug demand forecasting model reweights the training data to match the loss function of the machine learning model with the decision optimization goal, and assigns higher penalty weights to overestimated forecasts that result in drug expiration and waste, and underestimated forecasts that result in drug shortages.

[0058] The drug demand forecasting model is constructed based on an LSTM network of time series; the input features of the model include time features, trend features and external event features in the historical drug consumption data; the time features include month, season and holiday identifiers; the trend features include the moving average of the past 3 to 12 cycles; the external event features include epidemic indicators or seasonal disease epidemic indexes; the model training process adopts a weighted loss function, and the loss function of the machine learning model is matched with the decision optimization goal by reweighting the training data according to the degree of impact on the decision cost. A first penalty weight is assigned to the overestimation of the forecast that causes drug expiration and waste, and a second penalty weight is assigned to the underestimation of the forecast that causes drug shortages. The first penalty weight and the second penalty weight are greater than the weight of the standard sample, so that the model training goal is consistent with the optimization goal of the procurement decision; the drug demand forecasting model outputs the demand forecast value and its 95% confidence interval for the next 4 to 12 procurement cycles.

[0059] The penalty weight for overestimation is the unit expiration cost of drug i The penalty weight for underestimated prediction is the shortage cost. 3 times.

[0060] The following is a Python code example that uses an LSTM network to obtain a drug demand forecast model. Please note that this example is only a starting point and may need to be adjusted based on actual conditions and device interfaces.

[0061]

[0062]

[0063]

[0064]

[0065]

[0066]

[0067]

[0068]

[0069]

[0070] This code is only an example and needs to be modified and adjusted appropriately according to specific circumstances in actual applications.

[0071] Step S3: Generate the initial state of the current optimization cycle, including the estimated ending inventory of each drug if no purchase is made during this cycle and the quantity that may expire and be scrapped.

[0072] Step S4: construct a drug procurement optimization model and set constraints for the drug procurement optimization model.

[0073] The objective function of the optimization model is:

[0074]

[0075] Where, is the purchase cost coefficient of drug i, is the unit inventory holding cost coefficient, is the unit cost of expired goods (which can be equal to the drug cost or higher to reflect the implicit loss), is the unit shortage penalty cost (such as the replacement cost or potential treatment delay loss caused by drug shortage), x i,t is the quantity of drug i ordered in period t, I i,t is the effective quantity of drug i in inventory at the end of period t, E i,t is the number of expired and scrapped drugs i in period t, S i,t is the shortage quantity, T is the total number of planning cycles, and i∈I is any drug i in the drug set I.

[0076] The constraints of the drug procurement optimization model include:

[0077] Demand satisfaction constraints: For each cycle and drug, the sum of effective inventory and current purchases must meet demand, and the allowed unmet demand is counted as a shortage S i,t , and is penalized in the target, i.e. I i,t +x i,t +S i,t ≥D i,t , and S i,t ≥0, where D i,t To forecast demand;

[0078] Inventory evolution constraint: the inventory at the end of the current cycle is equal to the inventory of the previous cycle plus the current purchase minus the demand consumption, and negative inventory is not allowed. At the same time, it is subject to the storage capacity limit, that is, in is the storage capacity limit of drug i, (·) + Indicates that the lower limit is 0 (negative inventory is not allowed, that is, unmet demand can be counted as a shortage S i,t =maxD i,t -I i,t -x i,t ,0);

[0079] Inventory capacity constraint: the total inventory of all drugs in each cycle must not exceed the total capacity of the warehouse to prevent blind purchases that exceed the storage capacity, that is, in is the total warehouse capacity;

[0080] Procurement cycle and supply constraints: Drug orders can only be placed within the permitted procurement cycle, and the impact of supplier delivery lead time must be considered, i.e., x i,t It can only be taken within the allowed procurement cycle. If the supply lead time is cycle, then decide when making a decision Inventory replenishment after the cycle.

[0081] Other business constraints: such as procurement budget limits (purchase expenditure per cycle Not exceeding the budget), supplier batch restrictions (some drugs must be purchased in boxes at a time, such as x i,t is a specific multiple), etc., can be added to the model as needed.

[0082] Step S5: Divide the entire planning cycle into shorter rolling windows, solve the simplified sub-problem in each window and obtain the decision of the current cycle.

[0083] Step S6, executing the heuristic greedy and local search algorithms, including generating an initial feasible solution based on the reorder point strategy; making local adjustments to the drug order quantity in the initial solution, reducing the order quantity of drugs whose expected inventory exceeds the threshold, and increasing the order quantity of drugs whose expected shortage is greater than zero; under the premise of meeting the warehouse capacity limit, performing multiple rounds of iterative optimization until the improvement of the objective function value is less than the preset threshold or the maximum number of iterations is reached.

[0084] The heuristic greedy and local search algorithms include:

[0085] Initially, a feasible solution is generated based on empirical rules. Based on the (s, S) strategy, a reorder point and replenishment limit are set for each drug. If the inventory is lower than the reorder point, the drug is ordered to the replenishment limit.

[0086] Make local adjustments to the solution to reduce costs, identify drugs whose expected inventory after ordering exceeds a preset threshold, and try to reduce their order quantity to reduce expiration costs;

[0087] Increase order quantities for drugs with projected shortages greater than zero until the cost of the shortage penalty balances the cost of increased inventory;

[0088] When it is found that the total inventory exceeds the warehouse capacity limit, the order quantity of some drugs is reduced in ascending order according to the expiration date of the drugs so that the total inventory meets the capacity constraint.

[0089] Through multiple rounds of iteration, until the improvement of the objective function value is less than the preset threshold or the maximum number of iterations is reached.

[0090] Step S7: The solution generated by the optimization algorithm is presented on the purchasing decision support interface, listing the current inventory, predicted consumption, and recommended order quantity for each drug;

[0091] Step S8: Submit the confirmed order plan, generate a formal purchase order, and place the order with the supplier;

[0092] Step S9: After the supplier delivers the medicine and the medicine is put into storage, the inventory database is updated to register the date of entry and expiration date of the medicine;

[0093] Step S10, record the types and quantities of expired and scrapped drugs in the previous cycle, and adjust the drug demand forecast model and optimization parameters through the feedback mechanism.

[0094] A hospital easily expired drug procurement management system, comprising:

[0095] The data collection and status monitoring module is used to extract the latest inventory data and drug consumption data from the hospital information system. The inventory data includes the current inventory quantity of each drug, the time each batch entered the warehouse, and the warning information of the approaching expiration of the drugs in stock. The drug consumption data includes the recent records of the amount of drugs issued by each department. The module is also used to obtain parameters related to procurement management, including the purchase price of each drug, shelf life, supplier delivery cycle and batch requirements, remaining warehouse capacity and budget status;

[0096] A demand forecasting module is used to call a drug demand forecasting model to predict the demand for each drug in the next Δ period. The drug demand forecasting model reweights the training data to match the loss function of the machine learning model with the decision optimization goal;

[0097] The procurement optimization decision calculation module is used to generate the initial state of the current optimization cycle, build a drug procurement optimization model, divide the entire planning cycle into shorter rolling windows, and execute a heuristic greedy + local search algorithm to obtain the recommended ordering plan for the current cycle;

[0098] The decision-making, approval, and execution module is used to present the solution provided by the optimization algorithm, generate a formal purchase order, and connect with the supplier to place the order;

[0099] The inventory update and feedback learning module is used to update the inventory database after the supplier delivers the goods and verifies them into the warehouse, record the types and quantities of expired and scrapped medicines, and feed this data back to the demand forecasting module and procurement optimization decision calculation module to adjust the optimization algorithm parameters.

[0100] Example 1

[0101] In this embodiment of the present invention, refined procurement management is implemented for "Cefuroxime Sodium Injection (0.75g)." This drug is a commonly used clinical anti-infective medication with high frequency of use, making it susceptible to inventory expiration losses due to forecasting bias. The drug is purchased at 6.20 yuan per bottle, has an 18-month shelf life, and historical monthly usage ranges from 3,200 to 3,600 bottles. The hospital currently has 3,800 bottles in inventory, of which 680 will expire within the next 60 days. In step S2, the system first invokes a drug demand forecasting model based on a long-short-term memory neural network (LSTM). The model input features include historical drug consumption data, the number of inpatients with infections, seasonal factors, holiday distribution, and emergency room fluctuation indicators. The model predicts demand for 3,420 bottles in the next procurement cycle, with a confidence interval of 3,250 to 3,620 bottles. The training process utilizes a weighted loss function, assigning a 1.5x weight to overestimated predictions and a 2x weight to underestimated predictions, thereby enhancing sensitivity to decision-making outcomes.

[0102] The system then implemented a heuristic greedy and local search algorithm to solve the procurement strategy. Initially, the (s, S) strategy set the reorder point for the drug at 3,000 bottles, with an upper replenishment limit of 4,500 bottles. Based on current inventory and forecasted demand, the system initially recommended a purchase quantity of 700 bottles. During the local search process, the system considered parameters such as the unit expiration and scrapping cost (1.8 times the purchase price, or 11.16 yuan per bottle), the unit inventory holding cost (0.0775 yuan per bottle per month), and the shortage penalty cost (4 times the purchase price, or 24.8 yuan per bottle). The system determined that a purchase quantity of 700 bottles would increase the predicted expiration quantity to 130 bottles, which would be detrimental to overall cost control. After multiple rounds of iterative adjustments, the system ultimately recommended a purchase quantity of 500 bottles, reducing the expected number of expired drugs and limiting the risk of shortages. This recommendation was reviewed and implemented by the pharmacy department. After the procurement cycle ended, the system did not detect any new scrapped batches, the drug inventory turnover days dropped from 52 days to 39 days, and the inventory turnover rate increased to an average of 2.3 times per month, demonstrating the good practical effect of the method of the present invention in reducing the expiration rate and optimizing the inventory structure, as shown in Table 1.

[0103] Table 1

[0104]

[0105] Example 2

[0106] In the embodiment of the present invention, inventory optimization and feedback learning of "Gliclazide Sustained-Release Tablets (30 mg × 30 tablets)" are managed. This drug is an oral hypoglycemic drug for chronic diseases, with a purchase price of 12.5 yuan per box, a shelf life of 24 months, and an average monthly consumption of about 1,000 boxes. In the fourth quarter of 2023, the system discovered that a total of 132 boxes of this drug were scrapped due to high forecasts, resulting in direct economic losses of 1,650 yuan. In step S9 of the present invention, a total of 2,000 boxes of this drug were purchased in the first quarter of 2024. The system completed the warehousing of the supplier's delivery data and marked the expiration date of each batch. At the same time, a mapping table of the remaining available period of this batch of drugs was established in the near-expiry warning module. After the warehousing information is updated, the system starts the feedback mechanism in step S10 to analyze the reasons for scrapping, and identifies that the previous cycle's forecast model has a systematic overestimation in the context of a downward trend in outpatient volume during holidays. The system automatically adjusts the weights of training data, assigning a 1.7-fold weight to samples that cause expiration and scrapping, and a 1.3-fold weight to samples that are at risk of inventory expiring. At the same time, the unit scrapping cost of the drug set in the optimization model is increased from the original 13 yuan / box to 16 yuan / box to enhance the inhibitory effect on high inventory behavior.

[0107] After retraining the model and adjusting its parameters, the system regenerated procurement recommendations for the next two cycles, lowering the original recommended quantity from 2,000 boxes to 1,600 boxes. After purchasing staff adopted this recommendation, the hospital's total inventory decreased, and the inventory turnover cycle was shortened from 65 days to 42 days. No new scrap records were recorded within two months. Details are shown in Table 2.

[0108] Table 2

[0109]

[0110] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0111] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A hospital's method for optimizing procurement decisions for easily expired drugs, characterized by: The following steps are involved: Step S1, extracting the latest inventory data, drug consumption data and procurement management related parameters from the hospital information system; Step S2: calling the drug demand forecasting model to forecast the demand for each drug in the future cycle; Step S3: Generate the initial state of the current optimization cycle, including the estimated ending inventory of each drug if no purchases are made in this cycle and the predicted number of expired and scrapped drugs; Step S4: constructing a drug procurement optimization model and setting constraints for the drug procurement optimization model; Step S5: Divide the entire planning cycle into shorter rolling windows, solve the simplified subproblem in each window and obtain the decision of the current cycle; Step S6, executing a heuristic greedy and local search algorithm, including generating an initial feasible solution based on a reorder point strategy; locally adjusting the drug order quantity in the initial solution, and performing multiple rounds of iterative optimization until the improvement in the objective function value is less than a preset threshold or the maximum number of iterations is reached, while satisfying the warehouse capacity constraint; Step S7: The solution generated by the optimization algorithm is presented on the purchasing decision support interface, listing the current inventory, predicted consumption, and recommended order quantity for each drug; Step S8: Submit the confirmed order plan, generate a formal purchase order, and place the order with the supplier; Step S9: After the supplier delivers the medicine and the medicine is put into storage, the inventory database is updated to register the date of entry and expiration date of the medicine; Step S10, record the types and quantities of expired and scrapped drugs in the previous cycle, and adjust the drug demand forecast model and optimization parameters through the feedback mechanism.

2. A hospital expiring drug procurement decision optimization method according to claim 1, characterized in that: The inventory data includes the current inventory quantity of each drug, the time of entry of each batch, and the warning information of the expiration of the drugs in stock. The drug consumption data includes the recent records of the amount of drugs issued by the department; the procurement management related parameters include the purchase price of each drug, shelf life, supplier delivery cycle and batch requirements, warehouse remaining capacity and budget situation.

3. The method for optimizing hospital purchase decisions for easily expired drugs according to claim 1, characterized in that: The objective function of the drug procurement optimization model is: in, is the purchase cost coefficient of drug i, is the unit inventory holding cost coefficient, is the unit expiration and scrapping cost, is the unit shortage penalty cost, x i,t is the quantity of drug i ordered in period t, I i,t is the effective quantity of drug i in inventory at the end of period t, E i,t is the number of expired and scrapped drugs i in period t, S i,t is the shortage quantity, T is the total number of planning cycles, and i∈I is any drug i in the drug set I.

4. The method for optimizing hospital purchase decisions for easily expired drugs according to claim 1, characterized in that: The constraints of the procurement optimization model include: Demand satisfaction constraints: For each cycle and drug, the sum of effective inventory and current purchases must meet demand, and the allowed unmet demand is counted as a shortage S i,t , and is penalized in the target, i.e. I i,t +x i,t +S i,t ≥D i,t , and S i,t ≥0, where D i,t To forecast demand; Inventory evolution constraint: the inventory at the end of the current cycle is equal to the inventory of the previous cycle plus the current purchase minus the demand consumption, and negative inventory is not allowed. At the same time, it is subject to storage capacity constraints, that is, in is the storage capacity limit of drug i, (·) + Indicates that the lower limit is 0; Inventory capacity constraint: the total inventory of all drugs in each cycle must not exceed the total capacity of the warehouse, that is, in is the total warehouse capacity; Procurement cycle and supply constraints: Drug orders can only be placed within the permitted procurement cycle, and the impact of supplier delivery lead time must be considered, i.e., x i,t It can only take values ​​within the permitted procurement cycle. If the supply lead time is 1 cycle, the inventory replenishment after 1 cycle is determined during decision making.

5. The method for optimizing hospital procurement decisions for easily expired drugs according to claim 1, characterized in that: The heuristic greedy and local search algorithms include: Initially, a feasible solution is generated based on the empirical rule, and the reorder point s is set for each drug based on the (s, S) strategy. i and replenishment limit S i , if the inventory is lower than the reorder point, order to the replenishment limit; Make local adjustments to the solution to reduce costs, identify drugs whose expected inventory after ordering exceeds a preset threshold, and try to reduce their order quantity to reduce expiration costs; Increase order quantities for drugs with projected shortages greater than zero until the cost of the shortage penalty balances the cost of increased inventory; When it is found that the total inventory exceeds the warehouse capacity limit, the order quantity of some drugs is reduced in ascending order according to the expiration date of the drugs so that the total inventory meets the capacity constraint. Through multiple rounds of iteration, until the improvement of the objective function value is less than the preset threshold or the maximum number of iterations is reached.

6. The method for optimizing hospital purchase decisions for easily expired drugs according to claim 1, characterized in that: The drug demand forecasting model is constructed based on a long short-term memory neural network of time series; the input features of the model include time features, trend features and external event features in historical drug consumption data; the time features include month, season and holiday identifiers; the trend features include moving averages of the past 3 to 12 cycles; the external event features include epidemic indicators or seasonal disease epidemic indexes; the model training process adopts a weighted loss function, and the loss function of the machine learning model is matched with the decision optimization goal by reweighting the training data according to the degree of impact on the decision cost. A first penalty weight is assigned to the overestimation of the forecast that causes drug expiration and waste, and a second penalty weight is assigned to the underestimation of the forecast that causes drug shortages. The first penalty weight and the second penalty weight are greater than the weight of the standard sample; the drug demand forecasting model outputs the demand forecast value and its 95% confidence interval for the next 4 to 12 procurement cycles.

Images