Medical resource allocation method and system based on risk prediction distribution and stratified minimum guarantee constraint, storage medium and electronic device

Abstract

This invention belongs to the field of medical information processing technology and discloses a method, system, storage medium, and electronic device for allocating medical resources based on risk prediction distribution and tiered minimum guarantee constraints. This invention solves the resource lock-in problem caused by existing quota cap mechanisms by introducing a technical architecture that combines tiered minimum guarantee with dynamic cross-tier allocation. While ensuring minimum coverage for each risk tier, the system allows remaining quotas to compete globally across tiers, enabling the processor to dynamically allocate medical resources based on risk prediction data, thus improving the cost coverage efficiency of medical resources under budget constraints.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, and in particular to a method, system, storage medium, and electronic device for allocating medical resources based on risk prediction distribution and tiered minimum guarantee constraints. Background Technology

[0002] In healthcare management information systems, the allocation of medical resources to high-risk groups (such as chronic disease management programs and high-end medical resource allocation) typically faces strict budget constraints, meaning that intervention resources can only be provided to a very small percentage (e.g., 0.5%-10%) of the total population. Such allocation decisions require processing massive amounts of patient medical data and ensuring the representativeness and accuracy of the allocation results under the statistical characteristics of complex sampling designs.

[0003] Current healthcare resource allocation technologies primarily employ quota capping mechanisms or simple ranking methods based on a single risk score. Quota capping mechanisms rigidly lock the total budget to each subgroup according to stratification proportions, independently selecting predetermined quotas within each stratum. However, this rigid segmentation leads to inefficiency in healthcare resource allocation at stratification boundaries: when a stratum contains a large number of low-risk individuals while another stratum contains high-risk individuals, the budget is locked in the low-risk stratum, unable to be allocated across strata to high-value patients, resulting in systemic resource misallocation.

[0004] While ranking methods based on a single risk score (such as mean prediction or historical cost) can allocate across different levels, they lack technical means to address the heavy-tailed distribution characteristics of healthcare expenditures. Healthcare expenditure data exhibits typical heavy-tailed characteristics (the top 5% of the population bears approximately 50% of total expenditures). Existing technologies cannot effectively identify high-cost-risk groups under extremely tight budgets (such as 0.5%-1%), resulting in inefficient cost coverage of the allocation results.

[0005] Furthermore, existing allocation techniques lack auditable technical implementation methods. The allocation decision-making process often relies on human experience or black-box algorithms, making it impossible to generate machine-readable audit trail data, technically verify the robustness of allocation strategies, or convert statistical performance differences into equivalent budget quota quantification indicators that decision-makers can understand. This results in insufficient usability and credibility of decision support systems.

[0006] In view of this, the present invention is proposed. Summary of the Invention

[0007] The present invention aims to solve at least one of the above technical problems, and provides a method, system, storage medium and electronic device for allocating medical resources based on risk prediction distribution and hierarchical minimum guarantee constraints.

[0008] Compared with the prior art, the present invention has the following beneficial effects: This invention solves the resource lock-in problem caused by existing quota cap mechanisms by introducing a technical architecture that combines tiered minimum coverage with dynamic cross-tier allocation. While ensuring minimum coverage for each risk tier, the system allows remaining quotas to compete globally across tiers, enabling the processor to dynamically allocate medical resources based on risk prediction data, thus improving the efficiency of medical resource cost coverage under budget constraints.

[0009] This invention employs various digital representations of medical risks (including mean prediction, high-cost probability prediction, quantile prediction, and tail risk indicators) to comprehensively characterize the heavy-tailed distribution of medical expenditures. Under extremely tight budget conditions, this technical solution significantly improves the accuracy of identifying high-risk groups, enabling the allocation decision vector to capture more actual medical costs under the same budget constraints.

[0010] This invention implements budget capacity equivalence conversion, transforming the statistical performance differences of different allocation strategies into technical data on the equivalent number of allocation slots, generating machine-readable decision support information. This technique solves the problem of the disconnect between statistical results and decision-making language in existing technologies, improving the decision support efficiency and human-computer interaction efficiency of medical information processing systems.

[0011] This invention achieves full-process technical auditing of the medical resource allocation process by outputting complete audit trail information, including allocation decision vectors, cost coverage efficiency assessment results, strategy robustness verification results, and audit log data. This technology ensures the reproducibility and verifiability of allocation decisions, meeting the technical requirements of medical information systems for data integrity and decision credibility. Attached Figure Description

[0012] Figure 1 A flowchart illustrating the medical resource allocation method based on risk prediction distribution and hierarchical minimum guarantee constraint provided for embodiments of the present invention; Figure 2 A schematic diagram of the structure of a medical resource allocation system based on risk prediction distribution and hierarchical minimum guarantee constraint provided for an embodiment of the present invention; Figure 3 A schematic block diagram of an example electronic device provided in an embodiment of the present invention; Figure 4 A schematic diagram comparing the cost coverage efficiency of different medical resource allocation strategies under multiple budget levels, as provided in the embodiments of the present invention; Figure 5 A distribution diagram showing the difference between the paired Bootstrap strategies of the coverage lower limit strategy and the quota upper limit strategy provided in the embodiments of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be arbitrarily combined with each other. Those skilled in the art should understand that the following descriptions are merely exemplary means of implementing this invention and not limiting conditions; any technical means employed to achieve the same or similar technical effects as this invention should fall within the protection scope of this invention.

[0014] Figure 1 This is a flowchart illustrating a medical resource allocation method based on risk prediction distribution and tiered minimum guarantee constraints. (Refer to...) Figure 1 The first embodiment of the present invention provides a medical resource allocation method based on risk prediction distribution and hierarchical minimum guarantee constraints, which is applied to a medical information processing system and executed by the processor of the system, and includes the following steps: S101, Obtain medical expenditure risk prediction data and medical resource stratification control parameters for each individual in the candidate patient population. The medical expenditure risk prediction data includes at least two of the following: mean prediction data, high cost probability prediction data, quantile prediction data, or tail risk indicator data output by the medical risk prediction model. The medical resource stratification control parameters include stability markers and allocation granularity control markers. The stability markers are used to distinguish between stable risk stratification and micro-risk stratification. The allocation granularity control markers are used to specify stratified allocation or bucket-level allocation.

[0015] This step aims to provide multi-dimensional risk characterization inputs and tiered governance parameters for healthcare resource allocation decisions. From a technical implementation perspective, this step is typically executed by the data interface module of the healthcare information processing system, which receives output data from upstream healthcare risk prediction systems through standardized data exchange protocols.

[0016] The healthcare expenditure risk prediction data refers to the digital representation generated by quantitatively estimating the risk of an individual's future healthcare expenditure through machine learning models. This includes not only traditional point estimations (mean prediction), but more importantly, probability quantile predictions and tail risk indicators that characterize the heavy-tailed distribution of healthcare expenditures. The healthcare resource stratification control parameters refer to the governance metadata generated after stratifying high-risk candidate groups using clustering algorithms, used to constrain the granularity and stability requirements of subsequent allocation processes.

[0017] The specific implementation of data acquisition falls within the realm of conventional techniques in this field. The system can acquire patient medical data through data interfaces with regional public health information platforms, medical insurance settlement databases, or hospital information systems (HIS). In one specific embodiment, the system acquires annual patient medical utilization data from a medical big data platform. This data includes a 32-dimensional AB-expanded feature pattern (covering demographics, historical medical utilization, drug aggregation, and other features), as well as five types of digital representation data of medical risks output by an upstream prediction engine: mean prediction (MEAN), probability prediction of the top 5% high-cost patients (P...). TOP5 ), Top 1% high-cost probability prediction (P TOP1 The system employs five metrics: 95th percentile prediction (Q95), tail risk area index (TAIL), and 95th percentile prediction (Q95). These five metrics collectively constitute the prediction object library, enabling the system to select the most suitable ranking criteria under different budget constraints.

[0018] The medical resource stratification control parameters include a stability marker and an allocation granularity control marker. The stability marker distinguishes between stable-risk stratification and low-risk stratification. In one specific embodiment, when the proportion of the population in a stratum under complex sampling design weights (such as PERWT22F) is less than 5%, the stratum is determined to be a low-risk stratum; otherwise, it is a stable-risk stratification. The allocation granularity control marker specifies the smallest unit of allocation operation: stratum level or bucket level. For low-risk stratification, the system enforces bucket-level allocation to avoid statistical instability caused by overly fine granularity.

[0019] The complex sampling design weights refer to the design weights for complex sampling survey data, used to adjust for differences in sampling probabilities in statistical calculations to ensure the estimation of overall representativeness. Those skilled in the art should understand that for simple random sampling data, these weights can be set to an equal value of 1.0 without affecting the technical essence of subsequent allocation calculations.

[0020] It should be noted that although five predictive characteristics and a specific stratification threshold (5%) are listed in detail above, those skilled in the art should understand that the technical function of this step can be achieved as long as at least two different dimensions of risk prediction data (such as mean and quantiles) and control parameters that can distinguish stratification stability are provided. The specific selection of the 32-dimensional feature pattern and the five predictive characteristics is only one specific implementation method and is not a limitation on the scope of protection of this invention.

[0021] S102, Set a medical resource allocation budget, wherein the allocation budget determines the total number of patients who can receive medical resource allocation.

[0022] This step aims to determine the total capacity constraints for the allocation of available medical resources in the system, providing numerical boundaries for subsequent tiered protection and global allocation.

[0023] The medical resource allocation budget is represented in the computer system as the total number of patients K who can receive medical resource allocation, that is, the total number of patients determined by the allocation budget, which is composed of a predetermined proportion k% and the total number of candidate patients |C. t The product of | is determined, i.e., K = floor(k% × |C) t |).

[0024] Proportional calculation and rounding down are standard data processing techniques in this field. The system can flexibly set the k% value based on the resource capacity (or budget amount, project scale) of the actual medical management project. In one specific embodiment, to comprehensively evaluate the performance of the allocation strategy under different scarcity levels, the system scans at five budget levels: k% ∈ {0.5%, 1%, 2%, 5%, 10%}. This multi-budget level scanning mechanism helps identify the budgetary conditional frontier of strategy performance, i.e., different optimal strategies are applicable under extremely tight budgets (0.5%-1%) and relatively loose budgets (5%-10%). Those skilled in the art can use other numbers of budget levels (such as 3 or 7) or different proportion values, without affecting the technical essence of this step.

[0025] S103, based on the allocated budget, determine the minimum number of patients covered for each stable risk stratum. For each stable risk stratum, ensure that at least the minimum number of patients covered are selected in descending order of the medical expenditure risk prediction data to guarantee minimum medical resource coverage for that risk stratum. For micro-risk strata, medical resource coverage is guaranteed only at the bucket level.

[0026] This step aims to ensure minimum healthcare resource coverage for each risk stratum through technological constraints, preventing high-risk strata from being systematically neglected due to their smaller scale.

[0027] The minimum guaranteed number of patients refers to the minimum number of patients that the system forcibly allocates to each stable risk stratification, and its calculation formula is: K s floor =ceil(floor pct ×|H s |), where s represents the index identifier for stable risk stratification; K s floor This represents the minimum number of patients covered in the s-th stable risk stratification; ceil(·) represents the floor function, which rounds the result up to the nearest integer to ensure that the minimum number of covered patients is an integer; represents the preset minimum coverage percentage parameter (floor percentage), which is the minimum coverage threshold required within each stable stratification, and its value is a real number greater than 0 and less than 1; |H s| represents the number of candidate patients within the s-th stable risk stratification (|·| represents the cardinality operator for the set). For micro-risk stratification, the system only performs safeguards at the bucket level, meaning it does not enforce a specific number of individual slots, but rather ensures that the stratification as a whole is reflected in the allocation results.

[0028] Stratification and quota calculation are conventional algorithmic operations, but the differentiated guarantee strategy based on stability markers is the key technical means of this invention. The system first parses the stability markers in the stratification control parameters. For strata marked "stable," strict quota guarantees are implemented: according to the medical expenditure risk prediction data (such as MEAN or P...). TOP5 Sort the data in descending order and select the top K from that layer. s floor For each patient, the system skips the individual-level quota calculation for the micro-risk stratification marked as "microsegment_watch" and only marks the stratification as "bucket-level protection". The specific individuals in this stratification will compete with other strata for the remaining quota in the subsequent cross-stratification dynamic allocation steps.

[0029] In one specific embodiment, the system employs a 76-field medical resource allocation control instruction set output from the upstream stratification engine. This instruction set is a machine-readable structured data file (such as JSON or Parquet format), which explicitly records the stability flag field and the digital representation of cluster centers for each stratum. The system automatically performs the aforementioned determination by parsing the `stability_flag` field (with a value of "stable" to identify stable strata or "microsegment_watch" to identify micro strata) and the `segment_size_weighted` field (weighted population proportion value) in this instruction set. This determination method based on complex sampling design weights ensures the statistical representativeness of the stratification level and avoids sampling bias caused by simple headcount.

[0030] S104, the remaining slots after subtracting the minimum guaranteed slots for each tier from the allocated budget are selected across tiers in global descending order of the medical expenditure risk prediction data, so that high-risk patients are not restricted by tier boundaries.

[0031] This step aims to improve overall allocation efficiency through a cross-layer resource competition mechanism, and solve the technical problem of budgets being rigidly locked in low-value layers.

[0032] The remaining quota refers to the remaining allocable quota after deducting the sum of all stable tiered minimum guaranteed quotas from the total budget K, i.e., K. remain =K-Σ s K s floor , where K remainThe remaining available slots are for this purpose. The global descending cross-stratum selection means ignoring stratum boundaries and uniformly sorting all candidate patients (including remaining individuals within the stratum that have received minimum coverage and all individuals in the low-risk stratum) according to their risk prediction data, selecting them sequentially until K is filled. remain .

[0033] Sorting and selection operations are routine data processing, but the cross-stratification competition mechanism is the core technical means of this invention. Existing quota cap mechanisms lock remaining slots within each stratum according to stratification ratios, causing high-risk patients to be blocked by stratification boundaries. This step, by removing this rigid constraint, allows remaining slots to flow across strata, achieving dynamic optimization of medical resource allocation.

[0034] In one specific embodiment, the system maintains a global priority queue, where each element represents a candidate patient, and its key is the patient's medical expenditure risk prediction data (e.g., the mean prediction value when using the MEAN strategy, and the P value when using the P strategy). TOP5 The strategy uses the top 5% probability value. The system first performs a tiered minimum guarantee step, removing selected minimum guarantee individuals from each stable tier and marking them as "assigned"; then, it inserts the remaining individuals from each tier (including all individuals from the micro-risk tier) into the global queue; finally, it pops K individuals sequentially from the head of the queue. remain Individual patients are dynamically allocated across strata. This architecture ensures that high-value patients (regardless of their stratum) are always prioritized, significantly improving cost capture efficiency under budget constraints.

[0035] S105, output the medical resource allocation decision vector, the medical resource cost coverage efficiency evaluation results of each allocation strategy, the robustness verification results of the allocation strategy based on at least 1000 resampling, the budget capacity equivalence conversion results, and the audit log data of the medical resource allocation decision. The medical resource allocation decision vector is a binary decision vector, where each element represents whether the corresponding patient is selected to receive medical resource allocation, and under the same input conditions, the allocation decision vector is a deterministic output.

[0036] This step aims to generate machine-readable allocation decision and audit trail data to meet the technical requirements of medical information systems for decision reproducibility and verifiability, while converting statistical performance differences into quantitative indicators that decision-makers can intuitively understand.

[0037] The medical resource allocation decision vector is a binary decision vector x. i∈{0,1}, where each element represents whether the corresponding patient is selected to receive medical resource allocation (1 indicates selection, 0 indicates non-selection), and under the same input conditions, the allocation decision vector is a deterministic output (self-consistent ARI=1.000), ensuring that the same input will produce the same allocation result. The medical resource cost coverage efficiency refers to the proportion of the weighted sum of medical expenditures of the selected patient population to the total weighted sum of medical expenditures of the candidate patient population, where the weighting calculation adopts a complex sampling design weight. The budget capacity equivalent conversion result refers to the technical data that converts the cost coverage efficiency differences of different strategies into the equivalent number of allocation slots, where the equivalent number of allocation slots refers to the equivalent total number of patients required to achieve the same cost coverage efficiency.

[0038] As one specific implementation, the robustness verification results of the allocation strategy are obtained through a resampling-based statistical verification method, using at least 1000 resampling iterations to calculate the superiority probability and confidence interval of each strategy. The system performs multiple paired comparisons (e.g., pairwise comparisons covering 15 strategies at 5 budget levels), with each group evaluated for statistical significance through 1000 paired resampling iterations. Those skilled in the art can use other statistical verification methods (such as cross-validation, permutation tests, etc.) to achieve equivalent auditing results; the 1000 resampling iterations are merely one specific implementation method.

[0039] As one specific implementation method, the system uses the following formula to calculate the equivalent value of the allocated quota: K * =min{K':CC(Baseline Strategy,K')≥CC(Target Strategy,K)} where K represents the total number of patient slots determined by the allocation budget, K' represents the number of slots allocated under the baseline strategy, and K... * Let represent the number of equivalent allocated slots required for the target strategy to achieve the same medical resource cost coverage efficiency under the total number of patients, and min{•} denotes the minimum value function.

[0040] For example, if the target strategy (such as CFLOOR) achieves cost coverage efficiency at K=52 (5% budget), while the baseline strategy (MEAN) requires K'=60 to achieve the same, then the capacity equivalence ratio is K / K≈1.15, indicating that the target strategy is equivalent to saving approximately 15% of the budget slots. Those skilled in the art can use other equivalence conversion methods (such as calculations based on AUROC or net benefit differences), as long as the statistical performance difference can be converted into a resource quantification metric understandable to decision-makers.

[0041] As one specific implementation, the system employs a resampling-based statistical validation method to evaluate the robustness of the allocation strategy. The system performs 60 paired comparisons (covering pairwise comparisons of 15 strategies at 5 budget levels), with each group using 1000 resampling iterations to calculate the superiority probability and 95% confidence interval for each strategy. For example, when comparing the CFLOOR and CONTRACT strategies, the system calculates through resampling that CFLOOR significantly outperforms CONTRACT at medium-to-high budget levels (k≥2%) (approximately a 2.0-fold difference at k=5%, 95% CI [+0.029, +0.117]), while at low budget levels (k<2%), CFLOOR still outperforms CONTRACT, but the difference is not statistically significant. Those skilled in the art can use other statistical validation methods (such as cross-validation, permutation tests, etc.) to achieve equivalent audit results; the 1000 resampling iterations and 60 comparisons are merely one specific implementation method.

[0042] As one specific implementation, the system systematically evaluates 15 healthcare resource allocation strategies (Core-15 strategy family) across 5 budget levels, with a total of no less than 900 evaluations (15 strategies × 5 budgets × multiple indicators). The strategies include at least four categories: anchor family (random allocation, retrospective cost ranking), machine learning prediction family (ranking based on MEAN, P_TOP5, P_TOP1, Q95, TAIL), hierarchical ensemble family (quota cap CONTRACT, coverage cap CFLOOR), and stress testing family (theoretical optimal ORACLE). By comparing the cost coverage efficiency of these strategies, the system identifies the budget conditional frontier: under extremely tight budgets (k≤1%), machine learning prediction strategies significantly outperform retrospective cost ranking; under wider budgets (k≥2%), classical ranking methods remain competitive.

[0043] In some preferred embodiments, a robustness verification step is also included: repeating the allocation method on at least three different candidate patient set risk thresholds to assess the robustness of the allocation results to the candidate set definition parameters, and performing the allocation method on medical validation data independent of the training data source to verify the qualitative reproducibility of the allocation strategy.

[0044] In one specific implementation, the system repeatedly executes the above allocation method on at least three different candidate patient set risk thresholds (e.g., defining candidate sets based on predicted probabilities of entering the top 10% high-cost range, the top 5% high-cost range, and the top 1% high-cost range) to evaluate the robustness of the allocation results to the candidate set definition parameters. This ensures that the allocation decision does not fluctuate drastically due to minor adjustments to the upstream risk threshold.

[0045] In one specific implementation, the system executes the allocation method on medical validation data (such as medical insurance settlement data from different years or regional public health information platform data) independent of the training data source to verify the qualitative reproducibility of the allocation strategy. For example, the allocation results on the 2023 medical data are consistent in direction with the results on the 2022 validation data (i.e., the difference in CostCapture between the coverage lower bound strategy and the quota upper bound strategy is consistent, both showing that the coverage lower bound strategy is superior), indicating that the allocation strategy has cross-time stability.

[0046] The system's final output includes, but is not limited to: ① Medical resource allocation decision vector (binary file or database record); ② Medical resource cost coverage efficiency assessment results for each strategy (including confidence intervals); ③ Robustness verification results of allocation strategies (superiority probability); ④ Budget capacity equivalence conversion results (K... * ⑤ Audit log data for medical resource allocation decisions (including operation timestamps, strategy parameters, hierarchical markers, threshold settings, etc.). These outputs constitute a complete offline allocation audit system, supporting post-event traceability and pre-event planning for medical management decisions.

[0047] Those skilled in the art should understand that although this embodiment describes in detail the implementation of specific parameters such as MEPS data, five predictive representations, 15 strategy comparisons, and 1000 bootstrap iterations, these details are only to demonstrate the technical effects and feasibility of this patent. As long as the medical information processing system adopts a hierarchical minimum guarantee + cross-hierarchical dynamic allocation technical architecture, realizes the decision support function of budget capacity equivalence conversion, and outputs auditable allocation decision data, it falls within the protection scope of this invention, and is not limited to the specific data scale, number of strategies, or number of statistical verifications listed in this embodiment.

[0048] The first embodiment of the present invention will be further described below in conjunction with specific application scenarios.

[0049] Example 1: Budget Allocation for Managed Care Programs This example is applied to a chronic disease management project of a municipal medical insurance bureau, which requires selecting 0.5% of the 1 million insured individuals to be included in the annual key intervention list.

[0050] The system first acquires candidate medical risk prediction data from the regional public health information platform, including five characteristics such as mean prediction and the probability prediction of the top 5% high-cost risk, as well as a set of medical resource allocation control instructions from the upstream stratification system. The instruction set displays four risk stratifications: S1 (extreme risk, stable stratification, accounting for 8%), S2 (mainstream high risk, stable stratification, accounting for 40%), S3 (medium risk, stable stratification, accounting for 35%), and S4 (specialty drug intensive, stable stratification, accounting for 17%).

[0051] The allocation budget is set at K = 5000 people. The system implements a minimum guarantee for tiered allocation: based on a complex sampling design weight, the minimum guarantee is 42 people for tier S1 (ceil(0.05×840)), 200 people for tier S2, 175 people for tier S3, and 83 people for tier S4, totaling 500 people. The remaining 4500 slots are dynamically allocated across tiers: the system sorts all candidates in descending order based on the top 5% high-cost probability predictions, and selects the top 4500 people, regardless of their tier.

[0052] The final output includes an allocation decision vector (5000 binary labels), a cost coverage efficiency assessment result (CFLOOR strategy captures 14.22% of actual medical expenditures), and audit logs. Compared to a quota cap mechanism (where quotas are locked proportionally at each level, and remaining quotas do not compete across levels), this embodiment improves the cost capture rate by approximately 2 times under the same budget.

[0053] Figure 5 This is a bootstrap difference distribution chart for the coverage lower limit strategy and the quota upper limit strategy. The vertical axis is labeled in the format "Strategy 1 - Strategy 2," indicating that the cost coverage efficiency difference (CC Difference) between Strategy 1 and Strategy 2 is calculated with Strategy 2 as the benchmark. A positive difference (bar to the right / green) indicates that Strategy 1 is superior to Strategy 2; a negative difference (bar to the left / red) indicates that Strategy 2 is better. Figure 5 The paired Bootstrap differential distribution plot (95% confidence interval) shown visually verifies the above conclusion: at a representative budget level (e.g., k=5%), the differential distribution (CFLOOR-CONTRACT) between the coverage lower limit strategy and the quota upper limit strategy is completely to the right of zero (difference approximately +0.072, 95% CI does not include the zero point), indicating that the statistical validation results based on 1000 resamplings support the significant superiority of the coverage lower limit strategy (superiority probability 1.000).

[0054] Example 2: Multi-budget Level Decision Support System This embodiment is applied to the precision intervention platform of a commercial health insurance company, which needs to provide decision support for different budget scenarios.

[0055] The system constructs a Core-15 policy family, including: an anchor family (random assignment, retrospective cost ranking), a machine learning prediction family (ranking based on mean, top 5% probability, top 1% probability, 95th percentile, and tail area), a hierarchical ensemble family (quota cap, coverage cap), and a stress testing family (theoretical optimality). The system evaluates each policy at five budget levels (0.5%, 1%, 2%, 5%, 10%), with a total of 945 evaluations (15 policies × 5 budgets × multiple metrics).

[0056] Under a very tight budget of 0.5%, the system found that the strategy based on the prediction of the top 1% high-cost probability was significantly better than retrospective cost ranking (cost coverage efficiency difference +0.034, superiority probability 0.999); under a standard budget of 5%, the system found that the strategy based on mean prediction was significantly better than the clinical heuristic ranking strategy (cost coverage efficiency difference +0.061, 95% confidence interval [+0.015, +0.115], superiority probability 0.997); the difference between the mean prediction strategy and the retrospective cost ranking strategy was not significant (superiority probability 0.46), indicating that machine learning (ML) methods have a significant advantage over weak heuristics, but no statistical difference compared to ranking based on historical expenditures.

[0057] The output includes an audit panel containing a 15×5 policy performance matrix, generated by the system as follows: Figure 4 The budget conditional allocation frontier curves shown compare the cost coverage efficiency of the coverage lower bound strategy (CFLOOR), quota upper bound strategy (CONTRACT), retrospective cost ranking (COST), and mean forecast ranking (MEAN) at budget levels ranging from 0.5% to 10%. Figure 4 As shown, at a 5% budget level, the cost coverage efficiency of the CFLOOR strategy is significantly better than that of the CONTRACT strategy, which intuitively verifies the technical advantages of the coverage lower bound architecture.

[0058] Example 3: Fairness Audit Report This embodiment is applied to the scenario of fairness auditing in the allocation of resources within a regional medical consortium.

[0059] The system uses the FAIR (Fairness Constraint) strategy to allocate medical resources and calculates and outputs two key audit metrics: Selection Disparity (difference in medical resource selection rates among different population subgroups) and Min Bucket Coverage (minimum risk stratification coverage rate).

[0060] For example, the system calculates that the difference in selection rates between the urban employee medical insurance and the rural and urban resident medical insurance subgroups is 2.3%, and the minimum tiered coverage ratio is 8.4%, both of which meet the preset fairness audit thresholds. These indicators are output as bounded secondary indicators for medical management departments to assess the population subgroup balance of allocation decisions, but they are not used as constraints on allocation decisions; they are only used for ex-post audits and governance compliance checks.

[0061] Example 4: Strategy Comparison Audit Panel This embodiment is applied to the strategy effectiveness evaluation system of the medical insurance fund supervision department.

[0062] The system constructs a complete evaluation matrix containing 15 medical resource allocation strategies, and conducts systematic evaluations at five budget levels (0.5%, 1%, 2%, 5%, and 10%), with a total of no less than 900 evaluations. The 15 strategies include: anchoring strategy family (random allocation, retrospective cost ranking), machine learning prediction strategy family (ranking based on mean, top 5% probability, top 1% probability, 95th percentile, and tail area), hierarchical integration strategy family (quota cap constraint strategy, coverage lower limit strategy), and stress testing strategy family (theoretical optimal strategy).

[0063] System generated as Figure 4 The visualized audit panel shown compares the cost coverage efficiency of different healthcare resource allocation strategies across multiple budget levels. This chart, generated using a paired bootstrap method, demonstrates that the CFLOOR strategy remains statistically significant and optimal at medium to high budget levels (k≥2%), particularly showing a significant difference from the CONTRACT strategy at the 5% budget level (+7.3pp, p=1.000). At low budget levels (k<2%), it still performs optimally, but the difference is smaller. This provides decision-makers with an intuitive basis for strategy comparison.

[0064] System generated as Figure 5 The paired Bootstrap difference distribution plot shown illustrates the difference distribution and 95% confidence intervals of 1000 resampling comparisons of 15 strategies. Figure 5 As shown, the difference interval for the CFLOOR-CONTRACT comparison lies entirely in the positive region (approximately +0.072), while the MEAN-EXT2 comparison also shows a significant advantage. However, the confidence interval for the MEAN-COST comparison crosses zero, indicating that the difference is not significant. This visualization result corroborates the audit panel data, providing statistical robustness evidence for strategy selection.

[0065] Figure 2 This is a schematic diagram of the structure of a medical resource allocation system 200 based on risk prediction distribution and tiered minimum guarantee constraints, referenced from [reference]. Figure 2 The second embodiment of the present invention provides a medical resource allocation system 200 based on risk prediction distribution and tiered minimum guarantee constraints, including: an acquisition module 201, a setting module 202, a tiering module 203, an allocation module 204, and an output module 205. The functions of each module are described below: The acquisition module 201 is used to acquire medical expenditure risk prediction data and medical resource stratification control parameters for each individual in the candidate patient population. The medical expenditure risk prediction data includes at least two of the following: mean prediction data, high cost probability prediction data, quantile prediction data, or tail risk indicator data output by the medical risk prediction model. The medical resource stratification control parameters include stability markers and allocation granularity control markers. The stability markers are used to distinguish between stable risk stratification and micro-risk stratification. The allocation granularity control markers are used to specify stratified allocation or bucket-level allocation. Setting module 202 is used to set medical resource allocation budget K, wherein the allocation budget K determines the total number of patients who can accept medical resource allocation; The stratification module 203 is used to determine the minimum number of patients covered for each stable risk stratification based on the allocation budget K. For each stable risk stratification, patients are selected in descending order of the medical expenditure risk prediction data to ensure minimum medical resource coverage for that risk stratification. For micro-risk stratification, medical resource coverage is only guaranteed at the bucket level. The allocation module 204 is used to select the remaining quota after subtracting the minimum guaranteed quota for each stratum from the allocation budget K, and select them across strata according to the global descending order of the medical expenditure risk prediction data, so that high-risk patients are not restricted by the strata boundary. Output module 205 outputs the medical resource allocation decision vector, the medical resource cost coverage efficiency assessment results of each allocation strategy, the robustness verification results of the allocation strategy, the budget capacity equivalent conversion results, and the audit log data of the medical resource allocation decision.

[0066] Based on the above embodiments, the present invention also provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the steps in the medical resource allocation method based on risk prediction distribution and hierarchical minimum guarantee constraint in the first embodiment.

[0067] Figure 3 A schematic block diagram of an example electronic device 300 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0068] like Figure 3 As shown, the electronic device 300 may include a computing unit 301, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 302 or a computer program loaded from a storage unit 308 into a random access memory (RAM) 303. The RAM 303 may also store various programs and data required for the operation of the device 300. The computing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0069] Multiple components in device 300 are connected to I / O interface 305, including: input unit 306, such as keyboard, mouse, etc.; output unit 307, such as various types of monitors, speakers, etc.; storage unit 308, such as disk, optical disk, etc.; and communication unit 309, such as network card, modem, wireless transceiver, etc. Communication unit 309 allows device 300 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0070] The computing unit 301 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 301 performs the various methods and processes described above, such as a medical resource allocation method based on risk prediction distribution and stratified minimum guarantee constraints, or a model training method for medical resource allocation based on risk prediction distribution and stratified minimum guarantee constraints. For example, in some embodiments, the medical resource allocation method based on risk prediction distribution and stratified minimum guarantee constraints, or the model training method for medical resource allocation based on risk prediction distribution and stratified minimum guarantee constraints, can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 308. In some embodiments, part or all of the computer program can be loaded and / or installed on device 300 via ROM 302 and / or communication unit 309. When the computer program is loaded into RAM 303 and executed by computing unit 301, one or more steps of the medical resource allocation method based on risk prediction distribution and stratified minimum guarantee constraints or the model training method for medical resource allocation based on risk prediction distribution and stratified minimum guarantee constraints described above can be performed. Alternatively, in other embodiments, computing unit 301 can be configured by any other suitable means (e.g., by means of firmware) to execute the medical resource allocation method based on risk prediction distribution and stratified minimum guarantee constraints or the model training method for medical resource allocation based on risk prediction distribution and stratified minimum guarantee constraints.

[0071] Based on the above embodiments, the present invention also provides a non-transitory computer-readable storage medium storing computer instructions, the computer instructions being used to cause the computer to execute the medical resource allocation method based on risk prediction distribution and hierarchical minimum guarantee constraints disclosed in the embodiments of the present invention.

[0072] Based on the above embodiments, the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the medical resource allocation method based on risk prediction distribution and hierarchical minimum guarantee constraint disclosed in the embodiments of the present invention.

[0073] The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0074] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0075] In the several embodiments provided in this application, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or other forms.

[0076] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0077] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0078] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application.

[0079] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A medical resource allocation method based on risk prediction distribution and tiered minimum guarantee constraints, characterized in that, Applied to and executed by the processor of a medical information processing system, the method includes the following steps: Obtain medical expenditure risk prediction data and medical resource stratification control parameters for each individual in the candidate patient population. The medical expenditure risk prediction data includes at least two of the following: mean prediction data, high cost probability prediction data, quantile prediction data, or tail risk indicator data output by the medical risk prediction model. The medical resource stratification control parameters include stability markers and allocation granularity control markers. The stability markers are used to distinguish between stable risk stratification and micro-risk stratification, and the allocation granularity control markers are used to specify stratified allocation or bucket-level allocation. A medical resource allocation budget is set, which determines the total number of patients who can receive medical resource allocation; Based on the allocated budget, the minimum number of patients to be covered for each stable risk stratum is determined. For each stable risk stratum, patients are selected in descending order of the medical expenditure risk prediction data to ensure minimum medical resource coverage for that risk stratum. For micro-risk strata, medical resource coverage is only guaranteed at the bucket level. The remaining slots after subtracting the minimum guaranteed slots for each tier from the allocated budget are selected across tiers in global descending order of the medical expenditure risk prediction data, so that high-risk patients are not restricted by tier boundaries. The system outputs a medical resource allocation decision vector, the medical resource cost coverage efficiency assessment results of each allocation strategy, the robustness verification results of the allocation strategy based on at least 1000 resamplings, the budget capacity equivalence conversion results, and the audit log data of the medical resource allocation decision. The medical resource allocation decision vector is a binary decision vector, where each element represents whether the corresponding patient is selected to receive medical resource allocation, and the allocation decision vector is a deterministic output under the same input conditions.

2. The medical resource allocation method based on risk prediction distribution and tiered minimum guarantee constraint as described in claim 1, characterized in that, In the step of acquiring medical expenditure risk prediction data, the following five types of digital representation data of medical risks are acquired simultaneously: mean prediction data, top 5% high cost probability prediction data, top 1% high cost probability prediction data, 95th percentile prediction data, and tail risk area index data, forming a medical risk prediction object library.

3. The medical resource allocation method based on risk prediction distribution and tiered minimum guarantee constraint as described in claim 1, characterized in that, It also includes a multi-strategy comparison and evaluation step: systematically evaluate at least 15 medical resource allocation strategies at at least 5 budget levels, with a total of no less than 900 evaluations. The allocation strategies include random allocation strategy, retrospective medical cost ranking strategy, mean prediction ranking strategy, probability prediction ranking strategy, quantile prediction ranking strategy, quota upper limit constraint strategy, and the coverage lower limit strategy composed of the tiered minimum guarantee step and the cross-tiered dynamic allocation step. Among them, the 15 medical resource allocation strategies include at least the following four categories: the anchor strategy family, including random allocation strategy and retrospective medical cost ranking strategy; the machine learning prediction strategy family, including ranking strategies based on mean, probability, quantile and tail area; the hierarchical integration strategy family, including quota upper limit constraint strategy and coverage lower limit strategy; and the stress test strategy family, including theoretically optimal strategy.

4. The medical resource allocation method based on risk prediction distribution and tiered minimum guarantee constraint as described in claim 1, characterized in that, In the tiered minimum protection step, the criterion for determining micro-risk tiering is that the proportion of the population in the tier under the complex sampling design weight is less than 5%, and micro-risk tiering is only allowed to provide medical resource coverage at the bucket level.

5. The medical resource allocation method based on risk prediction distribution and tiered minimum guarantee constraint as described in claim 1, characterized in that, It also includes a healthcare allocation fairness audit step: calculating the differences in healthcare resource selection rates and the minimum risk stratification coverage ratios for each population subgroup, reporting them as bounded secondary indicators, but not as constraints on healthcare resource allocation decisions.

6. The medical resource allocation method based on risk prediction distribution and tiered minimum guarantee constraint as described in claim 1, characterized in that, The equivalent conversion result of the budget capacity is obtained as follows: using the mean prediction ranking strategy as the benchmark strategy, and through formula K... * =min{K':CC(Baseline Strategy,K')≥CC(Target Strategy,K)} calculates the equivalent value of allocation slots for other allocation strategies relative to the baseline strategy, where K represents the total number of patient slots determined by the allocation budget, K′ represents the number of allocation slots for the baseline strategy, and K... * Let represent the number of equivalent allocated slots required for the target strategy to achieve the same medical resource cost coverage efficiency under the total number of patients, and min{•} denotes the minimum value function.

7. The medical resource allocation method based on risk prediction distribution and tiered minimum guarantee constraint as described in claim 1, characterized in that, It also includes a robustness verification step: repeating the allocation method on at least three different candidate patient set risk thresholds to assess the robustness of the allocation results to the candidate set definition parameters; and / or The allocation method is executed on medical validation data independent of the training data source to verify the qualitative reproducibility of the allocation strategy.

8. A medical resource allocation system based on risk prediction distribution and tiered minimum guarantee constraints, characterized in that: include: The acquisition module is used to acquire medical expenditure risk prediction data and medical resource stratification control parameters for each individual in the candidate patient population. The medical expenditure risk prediction data includes at least two of the following: mean prediction data, high cost probability prediction data, quantile prediction data, or tail risk indicator data output by the medical risk prediction model. The medical resource stratification control parameters include stability markers and allocation granularity control markers. The stability markers are used to distinguish between stable risk stratification and micro-risk stratification, and the allocation granularity control markers are used to specify stratified allocation or bucket-level allocation. A setting module is used to set a medical resource allocation budget K, wherein the allocation budget K determines the total number of patients who can accept medical resource allocation; The tiered module is used to determine the minimum number of patients covered for each stable risk tier based on the allocated budget K. For each stable risk tier, patients are selected in descending order of the medical expenditure risk prediction data to ensure minimum medical resource coverage for that risk tier. For micro-risk tiers, medical resource coverage is only guaranteed at the bucket level. The allocation module is used to select the remaining slots after subtracting the minimum guaranteed slots for each tier from the allocation budget K, and then select them across tiers in global descending order of the medical expenditure risk prediction data, so that high-risk patients are not restricted by tier boundaries. The output module outputs the medical resource allocation decision vector, the medical resource cost coverage efficiency assessment results of each allocation strategy, the robustness verification results of the allocation strategy, the budget capacity equivalent conversion results, and the audit log data of the medical resource allocation decision.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the medical resource allocation method based on risk prediction distribution and hierarchical minimum guarantee constraint as described in any one of claims 1-7.

10. An electronic device, characterized in that, include: processor; A memory for storing a computer program; wherein, when the processor executes the computer program, it implements the steps of the medical resource allocation method based on risk prediction distribution and hierarchical minimum guarantee constraint as described in any one of claims 1-7.

Images